Multi-band intelligent antenna array beam forming method, device, equipment and medium

Abstract

The invention relates to the technical field of wireless communication, and provides a multi-band intelligent antenna array beam forming method and device, equipment and a medium. Dynamic frequency band switching is carried out according to multi-dimensional environment data to obtain frequency band decision parameters, three-dimensional beam modeling is carried out according to the frequency band decision parameters to obtain a beam weight matrix, and hybrid precoding is carried out on the beam weight matrix according to a coding mode to obtain a digital-analog precoding matrix. Performing temperature compensation correction on the digital-analog precoding matrix according to the array temperature distribution data to obtain an optimized precoding matrix, and performing federated learning optimization on a local beam weight model according to the optimized precoding matrix to obtain a global beam forming weight, and performing phase error iterative correction on the global beamforming weight according to the radiation field distribution data to obtain an antenna array control signal. According to the invention, through multi-stage and full-link optimization processing, a complex dynamic environment can be adapted, and a high-precision beam forming effect is ensured.

Description

Technical Field

[0001] The present application relates to the field of wireless communication technology, and in particular to a multi-band smart antenna array beamforming method, device, equipment and medium. Background Art

[0002] Multi-band antenna array beamforming control can respond to complex and changing channel environments in real time, achieve flexible switching between frequency bands, and suppress interference and multipath effects, thereby improving the coverage and signal quality of the communication system. By accurately controlling the beam direction and amplitude of each frequency band, it can effectively utilize spectrum resources, reduce interference risks, and adapt to the challenges of high-speed mobile scenarios and complex urban environments, ensuring the stability and efficiency of communication links.

[0003] Existing multi-band antenna array beamforming control methods are usually based on a fixed digital-analog hybrid precoding scheme, relying on the pre-designed phase and amplitude control strategy for each frequency band, obtaining the channel information of each user through static channel estimation, using the traditional iterative algorithm to solve the beam weight, and forming the desired beam through digital precoding and analog phase shifters. However, this method generally uses static presets and fixed models and parameters to separately process each frequency band, lacks real-time response to dynamic changes in the environment and adaptive optimization capabilities, and cannot fully adapt to temperature fluctuations, environmental interference, and instantaneous changes in complex multi-user scenarios. Summary of the invention

[0004] In view of this, the present application provides a multi-band smart antenna array beamforming method, apparatus, device and medium to solve the problem of insufficient real-time response capability.

[0005] A first aspect of the present application provides a multi-band smart antenna array beamforming method, the method comprising: Perform dynamic frequency band switching processing according to the detected multi-dimensional environmental data to obtain frequency band decision parameters; Performing three-dimensional beam modeling processing according to the frequency band decision parameters to obtain a beam weight matrix; Performing hybrid precoding processing on the beam weight matrix according to a preset coding method to obtain a digital-analog precoding matrix; Performing temperature compensation correction processing on the digital-analog precoding matrix according to the detected array temperature distribution data to obtain an optimized precoding matrix; Performing federated learning optimization processing on a preset local beam weight model according to the optimized precoding matrix to obtain a global beamforming weight; The global beamforming weights are subjected to iterative phase error correction processing according to the detected radiation field distribution data to obtain an antenna array control signal.

[0006] In an optional implementation, the performing dynamic frequency band switching processing according to the detected multi-dimensional environmental data to obtain the frequency band decision parameter includes: Perform parameter extraction and fusion processing on the detected multi-dimensional environmental data to obtain the environmental parameter matrix; The environment parameter matrix is ​​subjected to nonlinear mapping processing of a Gaussian kernel function by a preset support vector machine classification model to obtain a frequency band label and an initial frequency band decision parameter; Calculate the classification error according to the preset actual frequency band label and the frequency band label to obtain the classification error rate; According to a preset classification accuracy threshold and the classification error rate, an online learning update process is performed on the initial frequency band decision parameter to obtain the frequency band decision parameter.

[0007] In an optional implementation, performing three-dimensional beam modeling processing according to the frequency band decision parameters to obtain a beam weight matrix includes: Step S21, performing pilot signal transmission and reception processing according to the frequency band decision parameters by a preset channel estimator to obtain a multi-user channel state information matrix; Step S22, performing compressed sensing sparse reconstruction processing on the multi-user channel state information matrix to obtain an optimization objective function; Step S23, performing an iterative solution process of the optimization objective function using an alternating direction multiplier method to obtain an initial beam weight matrix; Step S24, performing Monte Carlo simulation processing on the initial beam weight matrix to obtain beam coverage data; Step S25, verifying the initial beam weight matrix according to a preset sidelobe level threshold and the beam coverage data, and updating the optimization objective function according to a preset method; Repeat step S23 to step S25 until the sidelobe level in the beam coverage data is less than the sidelobe level threshold, and use the initial beam weight matrix as the beam weight matrix.

[0008] In an optional implementation, the preset coding method includes an alternating optimization algorithm, and the performing hybrid precoding processing on the beam weight matrix according to the preset coding method to obtain a digital-analog precoding matrix includes: Performing singular value decomposition processing on the beam weight matrix to obtain an initial digital precoding matrix and an initial analog precoding matrix; Performing alternating optimization iterative processing on the initial digital precoding matrix and the initial analog precoding matrix according to the alternating optimization algorithm to obtain a converged and optimized analog precoding matrix; The analog precoding matrix is ​​subjected to a phase shifter controlled codeword quantization process to obtain the digital-analog precoding matrix.

[0009] In an optional implementation, performing temperature compensation correction processing on the digital-analog precoding matrix according to the detected array temperature distribution data to obtain an optimized precoding matrix includes: A thermal expansion coefficient temperature difference model is constructed for the detected array temperature distribution data according to the preset LTCC process characteristic data to obtain a temperature-phase error model; Performing phase compensation matrix construction processing on the digital-analog precoding matrix by using the temperature-phase error model to obtain a diagonal matrix; A matrix product compensation process is performed on the digital-analog precoding matrix according to the diagonal matrix to obtain the optimized precoding matrix.

[0010] In an optional implementation, performing a federated learning optimization process on a preset local beam weight model according to the optimized precoding matrix to obtain a global beamforming weight includes: Performing local loss function definition processing on the optimized precoding matrix to obtain interference power loss; Performing a stochastic gradient descent update process on a model parameter set in a preset local beam weight model according to the interference power loss to obtain a standardized beam weight model; The standardized beam weight model is subjected to a federal average aggregation process to obtain the global beamforming weight.

[0011] In an optional implementation, the performing iterative phase error correction processing on the global beamforming weight according to the detected radiation field distribution data to obtain the antenna array control signal includes: Performing acquisition signal preprocessing on the detected radiation field distribution data to obtain the radiation field vector; Performing antenna response matrix construction processing on the global beamforming weights to obtain an antenna array response matrix; Perform LM nonlinear iterative optimization processing according to the radiation field vector and the antenna array response matrix to obtain a phase error vector; A phase error compensation process is performed on the global beamforming weight according to the phase error vector to obtain the antenna array control signal.

[0012] A second aspect of the present application provides a multi-band smart antenna array beamforming device, the device comprising: A decision parameter module, used for performing dynamic frequency band switching processing according to the detected multi-dimensional environmental data to obtain frequency band decision parameters; A beam weight module, used for performing three-dimensional beam modeling processing according to the frequency band decision parameters to obtain a beam weight matrix; A hybrid coding module, used for performing hybrid precoding processing on the beam weight matrix according to a preset coding method to obtain a digital-analog precoding matrix; A temperature correction module, used for performing temperature compensation correction processing on the digital-analog precoding matrix according to the detected array temperature distribution data to obtain an optimized precoding matrix; A learning optimization module, used for performing a federated learning optimization process on a preset local beam weight model according to the optimized precoding matrix to obtain a global beamforming weight; The iterative correction module is used to perform phase error iterative correction processing on the global beamforming weight according to the detected radiation field distribution data to obtain an antenna array control signal.

[0013] A third aspect of the present application provides an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the multi-band smart antenna array beamforming method as described above when executing the computer program.

[0014] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the multi-band smart antenna array beamforming method as described above are implemented.

[0015] In summary, this application at least includes the following beneficial technical effects: 1. Combining accurate channel state acquisition, advanced compressed sensing and iterative optimization methods, it effectively reduces the sidelobe level and improves the mainlobe concentration, thereby achieving more precise beam control.

[0016] 2. The digital-analog hybrid precoding design not only reduces the implementation complexity, but also ensures the stability of beamforming under different working conditions through temperature compensation technology.

[0017] 3. Federated learning is used to achieve global collaborative optimization, which not only protects local data privacy, but also generates globally optimal beamforming weights under multi-node collaboration.

[0018] 4. Phase iterative correction based on radiation field distribution data enables the entire system to promptly correct deviations caused by hardware, temperature or environmental factors in actual deployment, ensuring signal transmission quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0020] Figure 1 It is a flow chart of a multi-band smart antenna array beamforming method provided in an embodiment of the present application; Figure 2 It is a functional module diagram of a multi-band smart antenna array beamforming device provided in an embodiment of the present application; Figure 3 It is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0021] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0022] like Figure 1 , which is a flow chart of a multi-band smart antenna array beamforming method provided in an embodiment of the present application. The multi-band smart antenna array beamforming method provided in an embodiment of the present application includes the following steps.

[0023] Step S1: Perform dynamic frequency band switching processing according to the detected multi-dimensional environmental data to obtain frequency band decision parameters.

[0024] In order to ensure that the environmental parameter data is accurate, reliable and representative, it is necessary to perform detailed parameter extraction and data fusion on the multi-dimensional environmental data collected by the RF front-end sensor array. Among them, the RF front-end sensor array can monitor a variety of environmental parameters in real time, such as signal strength, delay spread, Doppler frequency shift and other important information related to wireless signal transmission. The collection of these data is carried out simultaneously by multiple sensors, and each sensor is responsible for recording the environmental characteristic data at its location. After the data collection is completed, the various parameters collected by each sensor are pre-processed, filtered and standardized to eliminate noise and unnecessary interference, and the various data dimensions are fused. Through fusion processing, multi-source data are integrated together to form a unified environmental parameter matrix. Among them, the environmental parameter matrix is ​​used for subsequent frequency band decision processing to ensure the selection of the optimal working frequency band in different scenarios, so as to achieve efficient operation of the antenna array in different environments.

[0025] Specifically, the RF front-end sensor array collects the environmental data of the current scene in real time through the RF front-end, and records the parameters such as the signal strength, delay spread, and Doppler shift detected by each. ij Construct the original data set D={d ij}, where i represents the index of the sensor, j represents the index of the parameter dimension, and d ij represents the original parameter of the jth parameter collected by the i-th sensor. At the same time, for each original parameter d in the original data set D ij Noise filtering, normalization and standardization are performed. Common methods include mean normalization, maximum and minimum normalization or Z-score normalization. Finally, all standardized data are fused to construct an environmental parameter matrix E.

[0026] The collected raw parameters are preprocessed and standardized to eliminate noise and dimensionality effects. At the same time, the standardized data are integrated through data fusion so that each row represents the multidimensional parameters of a sensor and each column represents a specific parameter dimension. The obtained environmental parameter matrix E can accurately reflect the multidimensional characteristics of the current wireless environment.

[0027] Furthermore, the constructed environmental parameter matrix E is input into the preset support vector machine classification model. Among them, the support vector machine classification model is a classic supervised learning method, which can convert the input data into a high-dimensional feature space through nonlinear mapping to achieve effective classification of complex data. Since environmental data usually has nonlinear distribution characteristics, the Gaussian kernel function is used to perform nonlinear mapping on the data, and an optimal segmentation surface can be found in the high-dimensional space to classify the environmental data into different frequency band labels, such as "Sub-6GHz" or "millimeter wave". The frequency band label and the initial frequency band decision parameters can be obtained through nonlinear mapping processing, where the initial frequency band decision parameters include but are not limited to the center frequency f c And bandwidth BW. By mapping data from the original input space to a higher-dimensional feature space, effective classification of linearly inseparable problems can be achieved.

[0028] Specifically, the support vector machine classification model converts the multidimensional parameter data e of the i-th sensor in each row of the environmental parameter matrix E into i As a data point, the multidimensional parameter data e is transformed into iCompare with the support vector to achieve high-dimensional space mapping so that the originally nonlinearly separable data becomes linearly separable in the high-dimensional space. Then, the support vector machine classification model uses the constructed decision function to weight the mapping results of all support vectors and add the bias term to determine which frequency band the input environmental parameters should be classified into. Through nonlinear mapping and linear combination, the frequency band label and initial decision parameters are output.

[0029] Exemplarily, the support vector machine classification model determines the classification result through the decision function shown below: Among them, K(E,E i ) represents the RBF function. E represents the input environmental parameter matrix. E i The data vector representing the i-th support vector. represents the square of the Euclidean norm, which is used to measure the distance between two vectors. γ is the Gaussian kernel parameter, which is used to control the width of the feature distribution after mapping. exp( ) represents the natural exponential function. f(E) represents the decision function output obtained for the input vector E. α i Represents the weight coefficient corresponding to the i-th support vector. i Indicates the label of the i-th support vector, which takes a value of 0 or 1 (corresponding to different frequency bands, such as 0 for Sub-6GHz and 1 for millimeter wave). b represents the bias term. sign() represents the sign function, which outputs a positive value if the internal expression is greater than zero, otherwise a negative value is output to determine the final classification result.

[0030] Further, the support vector machine classification model determines the corresponding initial frequency band decision parameters according to the classification results and the preset mapping rules. In the embodiment of the present application, the support vector machine classification model determines the center frequency f in the initial frequency band decision parameters according to the classification results output by the decision function. c and bandwidth BW. When the classification result is less than 0, the Sub-6GHz band (i.e., center frequency f c =3.5GHz, bandwidth BW=100MHz); when the classification result is greater than or equal to 0, the millimeter wave frequency band is selected (i.e., the center frequency f c =28GHz, bandwidth BW=400MHz).

[0031] In order to evaluate the classification performance of the support vector machine model under the current environmental data, the classification error calculation is used to obtain the difference between the frequency band label and the preset actual frequency band label, and the classification error rate ε is determined accordingly. Among them, the actual frequency band label is usually provided by historical data or a real data set that has been manually verified as a reference standard for the model output. The calculation of the classification error rate ε can not only reflect the accuracy of the current model, but also provide a basis for subsequent online learning updates. By calculating the classification error rate ε, it can be determined whether the support vector machine classification model needs to be updated online to ensure that the accuracy of the frequency band decision parameters meets the preset requirements. If there is a large deviation between the output frequency band label and the actual label, it means that the generalization ability of the model in the current environment is insufficient and the model needs to be updated. By setting a specific classification accuracy threshold (for example, 5%), it is determined whether the initial frequency band decision parameters output by the support vector machine model meet the requirements. When the actual error exceeds the classification accuracy threshold, the system automatically triggers the online learning module to update the parameters of the support vector machine model, thereby iteratively updating the initial frequency band decision parameters; when the actual error does not exceed the classification accuracy threshold, the system determines the initial frequency band decision parameters as the final frequency band decision parameters.

[0032] In actual operations, each frequency band label and the actual frequency band label have only two values: 0 or 1 (where 0 represents Sub-6GHz and 1 represents millimeter wave). If the model prediction is completely correct, the absolute value of the difference between the frequency band label and the actual frequency band label is 0; if the model prediction is completely wrong, the absolute value of the difference between the frequency band label and the actual frequency band label is 1. Therefore, the classification error rate ε is between 0 and 1, and the smaller the value, the higher the model accuracy. Setting the expected classification accuracy threshold (such as 5%, that is, ε≤0.05) can be used as a basis for subsequent judgment on whether the model needs to be updated.

[0033] By comparing the deviation between the model output and the actual data, the model performance can be accurately quantified. When the classification error rate ε exceeds the preset threshold, it means that the current model has deviations in the classification of environmental data, which provides the necessary judgment criteria for online learning updates in subsequent steps.

[0034] Among them, when the classification error rate ε exceeds the threshold, online learning update is performed. The purpose of online learning update processing is to use the newly collected environmental data and actual frequency band labels to incrementally train the existing support vector machine model, thereby updating the support vector set and model parameters, and then regenerating more accurate initial frequency band decision parameters. Online learning update is a process of dynamically adjusting the model, which enables the system to adapt to real-time changes in the environment and continuously optimize the accuracy of frequency band decisions in long-term operation.

[0035] Specifically, new data will be added (i.e., the environmental parameters and actual labels obtained from the latest environmental perception) are added to the existing support vector set and recorded as the updated support vector set The optimization problem of the support vector machine model is solved again for the updated support vector set to update the model parameters. The objective function of the optimization problem is as follows: Among them, K(E i ,E j ) represents the similarity between support vectors i and j calculated by the Gaussian kernel function, and C is the penalty factor that controls the tolerance of misclassification.

[0036] Then, the frequency band label is calculated again through the decision function according to the updated support vector machine model, and new frequency band decision parameters are generated according to the established rules.

[0037] Step S2: Perform three-dimensional beam modeling processing according to the frequency band decision parameters to obtain a beam weight matrix.

[0038] The design of the pilot signal requires high autocorrelation and low cross-correlation characteristics to ensure that each channel can be accurately estimated in a multi-user environment. Therefore, in a wireless communication system, accurately obtaining channel state information is the key to achieving high-quality beamforming. At the same time, it should be understood that a channel estimator is preset in the antenna array control system to obtain channel information between multiple users based on the transmission and reception processing of the pilot signal. Specifically, first, the designed pilot signal is transmitted in the working frequency band (for example, Sub-6GHz or millimeter wave) selected according to the frequency band decision parameter through the RF front end; then, after the multi-user receiving device receives the pilot signal, the received signal is amplified, filtered and analog-to-digital converted to obtain discrete pilot signal data. The receiving module transmits the collected pilot signal data back to the channel estimator, which recovers the channel coefficients of each transmission link based on the known characteristics of the pilot signal through matched filtering, correlation operation and other technologies, thereby constructing a multi-user channel state information matrix.

[0039] In actual implementation, there is an independent channel between each transmitting antenna and each receiving antenna in the system. For example, in a system with Nt transmitting antennas and Nr receiving antennas, the pilot signal is transmitted to each receiving end through a wireless channel after being transmitted, and the data obtained by the receiving end is processed to form a complex matrix (i.e., a multi-user channel state information matrix) H∈C Nr×Nt . Each element h in the multi-user channel state information matrix ijRepresents the channel coefficient between the jth transmitting antenna and the ith receiving antenna, which contains amplitude and phase information. The multi-user channel state information matrix can not only reflect the fading of the channel, but also show complex characteristics such as multipath effects, thus providing an accurate basis for subsequent beamforming.

[0040] In the specific implementation process, the channel estimator uses the time domain or frequency domain estimation method to perform correlation matching according to the designed pilot signal sequence. The data after matched filtering will be processed by channel equalization to compensate for factors such as channel fading and noise. After the processing is completed, the channel estimator arranges the calculated channel coefficients into a multi-user channel state information matrix H according to the corresponding relationship between the receiving antenna and the transmitting antenna.

[0041] Furthermore, the obtained multi-user channel state information matrix H needs to be processed by compressed sensing. Compressed sensing is an efficient data reconstruction technology that is applicable to situations where channel coefficients are sparse. At the same time, since wireless channels usually have sparse characteristics in space, that is, most of the channel energy is concentrated in a few directions, while the energy in other directions is weak. The sparse reconstruction method can reduce the sampling requirements and reconstruct high-precision channel information.

[0042] During the implementation process, the multi-user channel state information matrix H is first mapped to a preset sparse basis (such as a discrete Fourier transform basis, a discrete cosine transform basis, etc.) to obtain a coefficient vector in which only a few components have significant values ​​and the remaining components are close to zero. Next, by constructing an optimization objective function, we strive to maintain sparsity while minimizing the reconstruction error. The embodiment of the present application uses an L1 norm regularization term to facilitate the acquisition of a sparse solution. Among them, the embodiment of the present application uses a compressed sensing algorithm to sparsely reconstruct the channel matrix and constructs the optimization objective function shown below: Where W represents the desired beam weight matrix, that is, the channel response can be reconstructed through this matrix. H represents the multi-user channel state information matrix. S represents the target signal spatial distribution matrix, which is usually a predefined ideal beam shape or target response. λ is a regularization parameter, which is used to balance the weight of reconstruction error and sparsity. Its selection is determined based on actual system requirements and empirical values, and the optimal value is generally obtained through cross-validation and other methods. The optimization objective function is used to solve a beam weight matrix W with sparse characteristics while minimizing the reconstruction error, so as to optimize the beamforming effect in subsequent steps. The first term of the optimization objective function It is used to measure the accuracy of reconstruction and ensure that the obtained beam weight matrix can make the reconstruction result as close as possible to the target response (i.e., the target signal spatial distribution matrix); optimize the second term of the objective function It is used to force the solution to have sparse characteristics and reduce unnecessary noise and redundant information.

[0043] Furthermore, the constructed optimization objective function is used to solve the beam weight matrix. The embodiment of the present application adopts the Alternating Direction Method of Multipliers (ADMM) to solve the beam weight matrix. The ADMM algorithm is a method suitable for solving large-scale optimization problems. Its core idea is to split the complex optimization problem into several sub-problems that are easier to solve, and through iterative updates, the solutions of each sub-problem gradually approach the global optimal solution of the original optimization problem. Specifically, the optimization objective function is first converted into a constrained optimization problem, and then the ADM algorithm is used to split it into two sub-problems, respectively dealing with the two-norm minimization and L1 norm regularization terms. The ADMM algorithm introduces Lagrange multipliers and penalty terms, and updates the beam weight matrix W and the auxiliary variable Z alternately, while updating the Lagrange multiplier U, so that the final beam weight matrix W can meet the original optimization goal. The whole process usually requires multiple iterations until the residual meets the preset convergence conditions.

[0044] The alternating direction multiplier method can split the optimization problem with non-smooth regularization terms (such as L1 norm) into tractable sub-problems. Optimize and then optimize the sparse regularization term ||Z|| 1 The Lagrange multiplier and the penalty parameter are adjusted to achieve a balance between the two in the iteration process. The ADMM algorithm has stability and convergence, and is often used in actual calculations. Its convergence speed is related to the selected penalty parameter ρ. It ensures that the initial beam weight matrix can present sparse characteristics while minimizing the reconstruction error, which is of great significance for the precise control of the subsequent beam shape. During the data processing process, each iteration calculates the current residual and checks whether the convergence condition is met (for example, the two norms of the residual are less than the predetermined threshold). If the convergence condition is not met, the iterative update continues until the convergence requirement is met. Throughout the process, the ADMM algorithm decomposes the complex non-smooth optimization problem into a series of simple matrix update steps, thereby achieving efficient solution.

[0045] After the initial beam weight matrix W is obtained, it needs to be processed by Monte Carlo simulation to verify its effect on spatial coverage. Monte Carlo simulation is a method of statistical analysis using random sampling technology to evaluate the performance of the system under various possible environments and random factors. The beam coverage data is obtained through simulation, especially focusing on the main lobe and side lobe characteristics of the beam to ensure that the beam design meets the requirements of signal gain and interference suppression in practical applications. In the specific implementation, the radiation field distribution of the antenna array in the target area is first simulated according to the initial beam weight matrix W. During the simulation process, random variables are introduced to simulate noise, multipath effects and other uncertain factors in the environment. The Monte Carlo method calculates the signal strength distribution of the beam in all directions under different conditions through a large number of random sampling. The simulation results are usually displayed in the form of beam coverage maps, signal gain distribution curves, etc., among which the sidelobe level data is particularly critical and is used to verify whether the weight matrix needs to be updated and optimized in the future.

[0046] Based on the beam weight matrix, the following formula is used to perform simulation calculations to obtain coverage data.

[0047] Where G(θ,φ) represents the beam gain in a certain direction in space (expressed by angles θ and φ). n Represents the complex weight corresponding to the nth antenna element in the beam weight matrix. Nt represents the number of transmitting antennas. is the phase factor, where k is the wave number and d n is the distance between the nth antenna element and the reference point. θ represents the elevation angle. φ represents the azimuth angle. By simulating the gain distribution of the antenna array synthetic beam in a given direction (θ, φ), a numerical description of the beam coverage is obtained. Through a large number of random sampling and calculations, the beam gain diagram in the entire space can be obtained, thereby evaluating the main lobe width and side lobe level.

[0048] Monte Carlo simulation provides a statistical analysis method for the evaluation of the initial beam weight matrix. Due to the large number of random interference factors in the wireless environment, a single numerical calculation cannot fully reflect the performance of the beam in the real environment. The introduction of Monte Carlo simulation can obtain statistical average results through repeated calculations, thereby more accurately reflecting the coverage and sidelobe characteristics of the beam. In this way, the problem of excessive sidelobe level in the design can be detected, and data basis can be provided for subsequent optimization steps. Each sampling in the simulation generates a beam coverage data, and after statistical analysis of all sampling results, a relatively stable coverage description can be obtained. This process ensures that the system design not only meets the theoretical requirements, but also can work stably in the actual random environment, thereby improving the robustness and reliability of the overall beamforming system.

[0049] To ensure that the sidelobe level of the beam is lower than the preset threshold, unnecessary interference is avoided in practical applications. The verification process mainly includes calculating the signal gain in each direction in the beam coverage data, focusing on the level of the sidelobe area. The preset sidelobe level threshold is usually determined according to the system design requirements (for example, the sidelobe level should be lower than the main lobe level by a certain number of decibels, such as -15dB). During the verification process of the embodiment of the present application, the simulation results are compared with the threshold. If the sidelobe level is detected to exceed the threshold, the optimization objective function needs to be updated. The updating method can be to adjust the regularization parameter λ to strengthen the sparse constraint and further suppress the sidelobe level. The updated optimization objective function is solved again by the ADMM algorithm to generate a new beam weight matrix, and the Monte Carlo simulation verification is repeated. This process forms a closed-loop optimization process until the sidelobe levels in all sampling results meet the design requirements. Finally, the initial beam weight matrix that meets the requirements is used as the final beam weight matrix for subsequent beamforming operations.

[0050] Specifically, the beam coverage data obtained through Monte Carlo simulation provides a large number of random sampling results, which can reflect the actual performance of the beam under different environmental conditions. By comparing with the sidelobe level threshold, it is possible to objectively judge whether the current design meets the standard. If it does not meet the standard, the sparsity requirement is strengthened by adjusting the regularization parameter in the optimization objective function to suppress the sidelobe. The updated optimization objective function will be solved again using the ADMM method to form a closed-loop feedback process until the sidelobe levels in all simulation results are lower than the preset threshold.

[0051] Step S3: perform hybrid precoding processing on the beam weight matrix according to a preset coding method to obtain a digital-analog precoding matrix.

[0052] It should be understood that the beam weight matrix contains the complex weighted information of each element of the antenna array in different beam directions. In order to convert the beam weight matrix into a precoding matrix that is convenient for the coordinated processing of the digital and analog parts, the singular value decomposition (SVD) method is used to decompose it. Among them, SVD is a mature and widely used mathematical method for matrix decomposition and dimensionality reduction. Its basic idea is to decompose an arbitrary matrix into the product of three matrices, including an orthogonal matrix, a diagonal matrix and another orthogonal matrix. In this way, the main characteristic components of the matrix are extracted, and the complex matrix operations are converted into a form with simpler structure and more efficient processing.

[0053] Specifically, the dimension of the beam weight matrix W is Nt×K (where Nt represents the number of transmit antennas and K represents the number of beams). First, the beam weight matrix W is decomposed into: is a left singular matrix whose column vectors form an orthogonal basis; It is a diagonal matrix, and the non-negative real numbers on its diagonal are called singular values, which reflect the energy distribution of the original matrix in all directions; is the conjugate transpose of the right singular matrix, and its row vectors also form an orthogonal basis. Secondly, the initial digital precoding matrix F is constructed based on the three matrices obtained by SVD decomposition BB (0) With the initial simulation precoding matrix F RF (0) Among them, the initial digital precoding matrix F BB (0) It is the product of the conjugate transpose of the diagonal matrix and the right singular matrix, which contains the main singular value information and the right singular vector; the initial simulated precoding matrix F RF (0) It is a left singular matrix, which retains the phase information and amplitude relationship of each unit in the antenna array. The complexity of the original beam weight matrix is ​​reduced to two parts, of which the digital precoding part mainly plays a role in baseband signal processing, while the analog precoding part performs phase control and signal amplification through RF hardware.

[0054] Since the initial digital precoding matrix F BB (0) With the initial simulation precoding matrix F RF (0) The optimal matching state has not yet been achieved, so further optimization is needed to make the two work together in the hybrid precoding system. The embodiment of the present application adopts an alternating optimization algorithm to optimize the initial digital precoding matrix F BB (0) With the initial simulation precoding matrix F RF (0) Perform alternating optimization iterative processing. The alternating optimization algorithm is an iterative method of step-by-step optimization. By alternately fixing one variable and optimizing another variable, it gradually approaches the global optimal solution. For the hybrid precoding problem, the optimization goal is to make the product of the digital precoding matrix and the analog precoding matrix as close as possible to the original beam weight matrix (that is, W≈F RF (k) F BB (k) ). Among them, F RF (k) and F BB (k) denote the digital and analog precoding matrices for the kth iteration, respectively.

[0055] Specifically, first fix F RF (k) Solving for F BB (k) , and then fix FBB (k+1) Solving for F RF (k+1) , until the error between the product of the two matrices and the beam weight matrix W meets the preset convergence condition. This process ensures that the analog precoding matrix obtained after multiple iterations can accurately reflect the modulation requirements of the antenna array signal, while the digital precoding matrix takes on the remaining digital processing. The specific iteration process is as follows: Fixed F RF (k) When , the digital precoding matrix is ​​obtained by pseudo-inverse calculation using the following formula: F BB (k+1) =F RF (k)† W Among them, F RF (k)† Indicates F RF (k) The pseudo-inverse.

[0056] Fixed F BB (k+1) When , the analog precoding matrix is ​​solved by gradient descent or other optimization methods using the following formula: Among them, ||*|| F Represents the Frobenius norm, which is used to measure the error between matrices.

[0057] Furthermore, after each alternating iterative optimization, (k+1) =||WF BB (k+1) F RF (k+1) || F 2 The error after each update is calculated and compared with the preset threshold until the error is less than the preset threshold or the maximum number of iterations is reached.

[0058] After completing the alternating optimization of the digital and analog precoding matrices, the analog precoding matrix F RF (k+1)It is close to the optimal state, but due to the discrete characteristics of the control of the phase shifter in the actual hardware implementation, it is necessary to convert the matrix of continuous values ​​into the corresponding control codeword. The embodiment of the present application maps each element in the analog precoding matrix to a preset set of phase shifter codewords, so that the digital-analog precoding matrix can be directly sent to the RF hardware for signal processing. Among them, the phase shifter usually supports a limited number of discrete phase controls. For example, each phase shifter may support 2, 4, 8 or 16 discrete phase states, which requires the optimized F RF (k+1) The codeword quantization process is performed so that each continuous phase value is mapped to the nearest discrete phase value.

[0059] Specifically, firstly, a preset phase shifter codebook is established, which lists all allowed phase values. RF (k+1) The phase part of each element in is extracted, and the distance between the phase value and each allowed phase value in the code book is calculated. The phase value with the smallest distance is selected as the mapping result. The quantized matrix is ​​the final digital-to-analog precoding matrix, which contains not only the digital precoding part F BB (k+1) , and also includes the analog precoding part F after discretization RF (k+1) , the two form a complete hybrid precoding matrix through matrix product.

[0060] Exemplarily, when the phase shifter supports 16 states, the codebook is set to the set C = {φ 1 ,……,φ 16}, the phase value is evenly distributed in the interval [0,2Π). RF (k+1) Each element f in mn Extract the corresponding phase θ mn =arg(f mn ), and then by Quantify the phase. mn Replace with , thereby obtaining the quantized analog precoding matrix F RF (quant) , and finally obtain the digital-analog precoding matrix GG=F RF (quant) F BB (k +1) .

[0061] The continuous analog precoding matrix is ​​quantized into hardware-implementable discrete phase control codewords (i.e., the digital-analog precoding matrix), ensuring that the precoding instructions sent to the RF module can be accurately executed by the phase shifter, while ensuring that the entire hybrid precoding matrix maintains a high performance level.

[0062] Step S4: performing temperature compensation correction processing on the digital-analog precoding matrix according to the detected array temperature distribution data to obtain an optimized precoding matrix.

[0063] In a multi-band smart antenna array system, temperature distribution has a direct impact on the physical structure and electrical performance of the antenna elements. Since the antenna array may be significantly affected by the ambient temperature during operation, the structural changes caused by temperature changes must be compensated to ensure the pointing accuracy of the beamforming. As an advanced antenna manufacturing process, the thermal expansion characteristics of the Low Temperature Co-fired Ceramic (LTCC) process determine that the phase response of the antenna element will shift during temperature changes. In order to compensate for this phase error caused by temperature changes, the detected array temperature distribution data is analyzed based on the preset LTCC process characteristic data, and a thermal expansion coefficient temperature difference model is constructed to obtain a temperature-phase error model.

[0064] In the actual implementation process, each antenna element in the array is integrated with a temperature sensor, which can collect temperature data of each part of the current array in real time. Nt×1 It is presented in matrix form, where Nt represents the number of antenna units. The preset LTCC process characteristic data provides information on the thermal expansion coefficient related to temperature changes. This coefficient is usually obtained through experimental measurement and reflects the proportion of the material's dimensional change when the temperature changes. The temperature difference model is a functional relationship between temperature changes and the phase error inside the antenna element. This relationship can be approximated as a linear or quadratic function. The temperature difference model can be expressed by the following formula: Δφ i =α(T i -T 0 )+β(T i -T 0 ) 2 Among them, Δφ i represents the phase shift of the ith antenna element due to temperature change. i represents the actual temperature detected by the ith antenna element. 0represents the reference temperature, which is generally taken as 25°C as the standard operating temperature. α is the linear thermal expansion coefficient, which represents the phase offset generated when the temperature rises by 1°C. β is the quadratic thermal expansion coefficient, which is used to describe the quadratic effect of temperature change. According to the degree to which the temperature of each antenna element deviates from the reference temperature, the phase error caused by thermal expansion is calculated, thereby providing a quantitative basis for subsequent phase compensation. Through this model, the impact of temperature changes can be quantified, so that the compensation system can automatically adjust the phase compensation parameters according to real-time temperature data.

[0065] After completing the construction of the temperature-phase error model, the model is used to perform phase compensation on the obtained digital-analog precoding matrix. Since the phase offset caused by temperature changes will have an adverse effect on the actual beamforming effect, the phase part in the digital-analog precoding matrix must be corrected. The goal of compensation is to construct a diagonal matrix C, in which the diagonal elements are phase compensation factors, which are used to offset the phase error caused by temperature changes, so that the precoding matrix can maintain the predetermined beam pointing accuracy after correction. Specifically, each element in the temperature-phase error vector Δφ is obtained according to the temperature-phase error model, and a corresponding compensation factor is constructed. For the i-th antenna element, its compensation factor is e -jΔφi Arrange the compensation factors of all antenna elements into a diagonal matrix C = diag(e -jΔφ1 ,……,e -jΔφNt ), where only the elements on the diagonal of the diagonal matrix C are non-zero, and the rest of the elements are zero. The construction process of the diagonal matrix C makes full use of the properties of the complex exponential function, and the positive phase offset caused by temperature is offset by the negative sign, so that after the precoding matrix is ​​corrected by matrix multiplication, the overall phase deviation is eliminated.

[0066] Finally, the digital-analog precoding matrix GG and the diagonal matrix C are combined to eliminate the phase deviation caused by temperature changes and obtain the final optimized precoding matrix. Specifically, the diagonal matrix C is multiplied on the digital-analog precoding matrix GG to adjust the phase of each antenna unit in the entire precoding matrix to restore it to the expected state. The obtained HH=C×GG represents the optimized precoding matrix after temperature compensation correction, which is used to describe that the phase information of each antenna unit in the digital-analog precoding matrix GG will be adjusted by the corresponding compensation factor in the diagonal matrix C, thereby eliminating the offset caused by temperature.

[0067] Step S5: performing federated learning optimization processing on a preset local beam weight model according to the optimized precoding matrix to obtain a global beamforming weight.

[0068] It should be understood that in federated learning optimization, each local node (for example, each base station) locally optimizes the precoding matrix according to the current environment and user conditions. First, a loss function is defined locally to quantify the interference rate loss and reflect the gap between the current precoding matrix and the ideal shaping effect. Among them, the loss function is used to define a measure based on interference power. In a multi-user environment, each user's received signal will be interfered with by signals from other users. For the precoding matrix, by adjusting its weights, the interference caused by non-target user signals can be suppressed to a certain extent. To this end, when processing the precoding matrix locally, it is necessary to first define a local loss function, which takes the sum of multi-user interference powers as the objective function.

[0069] For example, there are K local users, and the channel vector of each user is h k (where k=1,2,…,K), the target signal vector S k Represents the direction in which the signal enhancement is expected to be achieved at the user. In the design of the local loss function, the interference power loss can be defined as the sum of the non-target signal interference power between all users, so the local loss function can be expressed by the following formula: Where L(W) represents the loss function defined for the optimized precoding matrix W, reflecting the overall interference power loss. K represents the number of users in the local processing environment. k H represents the conjugate transpose of the kth user channel vector, capturing the channel characteristics of the user. W represents the current optimized precoding matrix, which contains the weighting information of each antenna element. S j Represents the target signal vector of the jth user, which is used to define the signal enhancement direction. The overall interference loss is quantified by calculating the interference power of each user when receiving non-target user signals and then adding the interference power of all users. The lower the value of this loss function, the better the effect of the precoding matrix on interference suppression, and the more the shaping effect meets the design goal.

[0070] Furthermore, according to the loss function L(W), the parameters of the local beam weight model are updated by stochastic gradient descent, so as to optimize the model and obtain a standardized beam weight model. Among them, stochastic gradient descent is a classic optimization algorithm used to quickly find the local optimal solution of the objective function on a large-scale data set. By continuously iteratively updating the model parameters, the loss function L(W) is minimized, thereby reducing the interference power loss of the precoding matrix and achieving an optimization effect. Among them, the local beam weight model includes a set of parameters θ, which include weighting factors and phase adjustment amounts of each antenna unit, and their initial values ​​are usually determined by previous precoding processing or historical data. In each iteration, the interference power loss calculated by the current loss function is used to update the model parameters according to the gradient descent method.

[0071] Specifically, first, according to the loss function L(W) and the model parameter set θ in the current local beam weight model (t) The gradient is calculated; then, the gradient is multiplied by the preset learning rate η to obtain the update amount, and the update amount is subtracted from the current parameter to obtain the updated parameter θ (t+1) . Among them, η represents the learning rate, which is a positive real number that determines the step size of each parameter update. The stochastic gradient descent update process is iterated repeatedly in the local node until the loss function value converges, thereby obtaining a standardized beam weight model. Among them, the criteria for the convergence of the loss function value mainly include the following three items: 1. Judgment of absolute change: By calculating the difference between the loss function values ​​of two adjacent iterations, and when the difference in the loss function value is less than the preset threshold, it is considered to have converged.

[0072] 2. Relative change judgment: By calculating the normalized absolute change of the loss function value between two adjacent iterations, and when the normalized absolute change is lower than the preset percentage, it is considered to have converged.

[0073] 3. Maximum number of iterations: If the number of iterations reaches the preset maximum value, the update will be stopped even if the convergence conditions are not fully met to ensure that the algorithm ends under limited resources.

[0074] After completing the local beam weight model parameter update, each local node (e.g., each base station) obtains a parameter vector θ in the standardized beam weight model. i *, where i represents the i-th node. In the federated learning framework, in order to ensure that the overall system works collaboratively in a multi-user, multi-base station scenario, each node needs to aggregate the local model parameters to form a global beamforming weight. The embodiment of the present application adopts federal average aggregation processing to obtain global beamforming weights, wherein federal average aggregation processing is the most commonly used model parameter integration method in federated learning. The core idea is to take the arithmetic mean of all locally updated model parameters to form a globally unified model, which can take into account the local characteristics of each node and improve the overall system performance.

[0075] Specifically, in actual data processing, each base station or node independently updates a parameter vector θ in a standardized beam weight model according to its own environment and user conditions. i * Then, by summing up all these model parameters and averaging the sum, a global beamforming weight Θ can be obtained. The arithmetic average process ensures that the contribution of each local model is taken into account, thereby achieving the purpose of balancing the differences between nodes. During the data processing process, each node only transmits its model parameters to the central coordination node or directly calculates through the distributed aggregation algorithm, without the need to transmit the original data, thereby protecting user privacy.

[0076] Step S6: performing phase error iterative correction processing on the global beamforming weights according to the detected radiation field distribution data to obtain an antenna array control signal.

[0077] In the antenna array system, accurately capturing the actual radiation field distribution data is the basis for achieving precise beam control. The system collects the radiation signals in the environment in real time by deploying a near-field probe array near the antenna array. The collected signals are usually affected by multipath effects, noise and environmental interference. Directly using these raw data will reduce the accuracy of subsequent processing. Therefore, the first step is to pre-process the collected signals, extract effective information, and form a radiation field vector. Specifically, first, each near-field probe obtains the radiation field signal data according to a predetermined sampling frequency. Among them, the original signal output by each probe is p i (t), where i = 1, ..., L represents the probe number, L represents the total number of probes, and t represents the time variable. Secondly, a digital filter (such as a low-pass filter or a band-pass filter) is used to filter the collected original signal p i (t) is processed to remove high-frequency noise and non-target frequency band interference to obtain the signal p after filtering and noise reduction i filt (t). For signal p i filt (t) Perform fast Fourier transform to convert the time domain signal into frequency domain representation to obtain the frequency domain signal p i(f) and extract the amplitude and phase information. i (f) Extract the amplitude of the target frequency band |p i (f 0 )| and phase ∠p i (f 0 ) information, where f 0 is the target operating frequency. Finally, the complex signals (comprehensive amplitude and phase information) extracted by all probes at the target frequency are arranged in a certain order to form a radiation field vector Each element in the radiation field vector reflects the complex value of the radiation field of the corresponding probe at the target frequency, representing the local field strength and phase information.

[0078] At the same time, in order to apply the global beamforming weight Θ to the actual radiation field correction, it is necessary to combine it with the physical structure of the antenna array to construct the antenna array response matrix. This matrix describes the response characteristics of each antenna unit in the array to different spatial angles, and is usually constructed based on information such as the geometric layout of the antenna array, the distance between units, and the operating frequency. Specifically, a uniform linear array or uniform planar array is first used to calculate the propagation wavelength and the displacement vector of the relative reference point between each antenna unit based on the known antenna array geometry (that is, the relative position of each antenna unit) and the operating frequency. Among them, any given direction is described by the elevation angle θ and the azimuth angle φ, so the response of a single antenna unit can be expressed as , where d n represents the distance of the displacement projection of the nth antenna unit in the wave propagation direction, and k=2Π / λ is the wave number. Finally, the responses of all antenna units in each desired direction are arranged to form the antenna array response matrix AΘ. The i-th row in the antenna array response matrix AΘ corresponds to the i-th probe measurement direction, reflecting the response characteristics of the entire antenna array in this direction. The antenna array response matrix AΘ is constructed based on antenna array theory and phase response principles, and can describe the overall response of the antenna array to any given direction.

[0079] After obtaining the radiation field vector P and the antenna array response matrix AΘ, the global beamforming weights are accurately corrected for phase errors based on these data. Since there is a deviation between the actual radiation field and the ideal antenna response, the Levenberg-Marquardt (LM) algorithm is used for nonlinear iterative optimization to solve the phase error vector of each antenna element, thereby providing a quantitative basis for the final correction. Among them, the LM algorithm is an iterative algorithm for solving nonlinear least squares problems. Its core is to perform Taylor expansion on the objective function and dynamically switch between the Newton method and the gradient descent method, thereby ensuring both a faster convergence speed and stability in nonlinear problems.

[0080] Specifically, the objective function F(Δθ) is defined according to the radiation field vector P and the antenna array response matrix Aθ, where Δθ∈R Nt×1 Represents the phase error vector to be solved. The objective function F(Δθ) is used to measure the error between the output of the antenna array response matrix after correction and the actual radiation field vector. The objective function can be expressed by the following formula: Where Δθ represents the phase error vector to be solved, with a size of Nt×1. P represents the radiation field vector, with a size of L×1. AΘ represents the antenna array response matrix, with a size of L×Nt. jΔθ represents applying a complex exponential operation to each element in Δθ to convert it into a phase compensation factor.

[0081] Furthermore, through the iterative update formula of the LM algorithm, Δθ is continuously adjusted to minimize the objective function F(Δθ). Initially, Δθ is set (0) = 0. The iterative update formula can be expressed as follows: Where r(k)=P-AΘe jΔθ is the residual vector. J is the Jacobian matrix of the objective function for Δθ. μ is the damping factor, which is used to adjust the balance between the Newton method and the gradient descent method in the LM algorithm. I is the identity matrix.

[0082] The iterative process continues until ||r (k) || 2 is less than the preset threshold or reaches the maximum number of iterations, thereby obtaining the final phase error vector Δθ * .

[0083] After obtaining the phase error vector Δθ * Afterwards, the vector is used to perform phase error compensation on the global beamforming weight Θ, thereby generating the final antenna array control signal. Due to the phase offset error in practical applications, the global beamforming weight Θ must be corrected to ensure that the beam pointing output by the antenna array is consistent with the design target. The embodiment of the present application adopts a simple and direct phase compensation method, that is, a complex phase correction factor is applied to each element in the global weight, and its value is , where Δθ i * is the compensation phase value of the ith antenna element. These compensation factors are multiplied by the global beamforming weights to form the final antenna array control signal matrix.

[0084] Specifically, assume that the global beamforming weights are vector Θ∈C Nt×1, where each element contains the amplitude and initial phase information. First, use the phase error vector Δθ * ∈R Nt×1 , calculate the compensation factor for each antenna element , forming the phase compensation factor vector C phase ∈C Nt×1 Finally, the global beamforming weights and the compensation factor vector are corrected by element-by-element multiplication to obtain the corrected antenna array control signal vector JJ∈C Nt×1 The correction process ensures that the phase part of the precoding weight of each antenna unit is corrected accordingly when the actual signal is transmitted, thereby offsetting the phase error caused by temperature, hardware inconsistency or environmental changes. The final generated antenna array control signal JJ will be directly sent to the antenna array control unit to achieve high-precision, low-error beamforming output.

[0085] The present application is applied to the field of wireless communication technology, by dynamically switching the frequency band of a multi-band antenna array according to multi-dimensional environmental data to obtain frequency band decision parameters, performing three-dimensional beam modeling according to the frequency band decision parameters to obtain a beam weight matrix, performing hybrid precoding on the beam weight matrix according to the coding method to obtain a digital-analog precoding matrix, performing temperature compensation correction on the digital-analog precoding matrix according to the array temperature distribution data to obtain an optimized precoding matrix, performing federated learning optimization on the optimized precoding matrix through a local beam weight model to obtain a global beamforming weight, and performing phase error iterative correction on the global beamforming weight according to the radiation field distribution data to obtain an antenna array control signal. The present application has achieved breakthroughs in various links such as frequency band selection, beamforming, temperature compensation, and global collaborative optimization through multi-stage, full-link optimization processing. It can not only adapt to complex dynamic environments, but also ensure high-precision beamforming effects under actual hardware conditions, thereby greatly improving the performance and reliability of wireless communication systems.

[0086] like Figure 2 , which is a functional module diagram of a multi-band smart antenna array beamforming device provided in an embodiment of the present application.

[0087] In some embodiments, the multi-band smart antenna array beamforming device 2 may include multiple functional modules composed of computer program segments. The computer programs of the various program segments in the multi-band smart antenna array beamforming device 2 may be stored in a memory of a server and executed by at least one processor to perform (see Figure 1 Description) Functionality of multi-band smart antenna array beamforming method.

[0088] In this embodiment, the multi-band smart antenna array beamforming device 2 can be divided into multiple functional modules according to the functions it performs. The functional modules may include: a decision parameter module 21, a beam weight module 22, a hybrid coding module 23, a temperature correction module 24, a learning optimization module 25, and an iterative correction module 26. The module referred to in the present invention refers to a series of computer program segments that can be executed by at least one processor and can complete fixed functions, which are stored in a memory. In this embodiment, the functions of each module will be described in detail in subsequent embodiments.

[0089] The decision parameter module 21 is used to perform dynamic frequency band switching processing according to the detected multi-dimensional environmental data to obtain frequency band decision parameters.

[0090] In an optional implementation, the decision parameter module 21 is specifically used to: Performing parameter extraction and fusion processing on the multi-dimensional environmental data to obtain an environmental parameter matrix; The environment parameter matrix is ​​subjected to nonlinear mapping processing of a Gaussian kernel function by a preset support vector machine classification model to obtain a frequency band label and an initial frequency band decision parameter; Calculate the classification error according to the preset actual frequency band label and the frequency band label to obtain the classification error rate; According to a preset classification accuracy threshold and the classification error rate, an online learning update process is performed on the initial frequency band decision parameter to obtain the frequency band decision parameter.

[0091] The beam weight module 22 is used to perform three-dimensional beam modeling processing according to the frequency band decision parameters to obtain a beam weight matrix.

[0092] In an optional implementation, the beam weight module 22 is specifically configured to: Step S21, performing pilot signal transmission and reception processing according to the frequency band decision parameters by a preset channel estimator to obtain a multi-user channel state information matrix; Step S22, performing compressed sensing sparse reconstruction processing on the multi-user channel state information matrix to obtain an optimization objective function; Step S23, performing an iterative solution process of the optimization objective function using an alternating direction multiplier method to obtain an initial beam weight matrix; Step S24, performing Monte Carlo simulation processing on the initial beam weight matrix to obtain beam coverage data; Step S25, verifying the initial beam weight matrix according to a preset sidelobe level threshold and the beam coverage data, and updating the optimization objective function according to a preset method; Repeat step S23 to step S25 until the sidelobe level in the beam coverage data is less than the sidelobe level threshold, and use the initial beam weight matrix as the beam weight matrix.

[0093] The hybrid coding module 23 is used to perform hybrid precoding processing on the beam weight matrix according to a preset coding method to obtain a digital-analog precoding matrix.

[0094] In an optional implementation, the hybrid coding module 23 is specifically used for: Performing singular value decomposition processing on the beam weight matrix to obtain an initial digital precoding matrix and an initial analog precoding matrix; Performing alternating optimization iterative processing on the initial digital precoding matrix and the initial analog precoding matrix according to a preset alternating optimization algorithm to obtain a converged and optimized analog precoding matrix; The analog precoding matrix is ​​subjected to a phase shifter controlled codeword quantization process to obtain the digital-analog precoding matrix.

[0095] The temperature correction module 24 is used to perform temperature compensation correction processing on the digital-analog precoding matrix according to the detected array temperature distribution data to obtain an optimized precoding matrix.

[0096] In an optional embodiment, the temperature correction module 24 is specifically used for: Constructing a thermal expansion coefficient temperature difference model for the array temperature distribution data according to preset LTCC process characteristic data to obtain a temperature-phase error model; Performing phase compensation matrix construction processing on the digital-analog precoding matrix by using the temperature-phase error model to obtain a diagonal matrix; A matrix product compensation process is performed on the digital-analog precoding matrix according to the diagonal matrix to obtain the optimized precoding matrix.

[0097] The learning optimization module 25 is used to perform a federated learning optimization process on a preset local beam weight model according to the optimized precoding matrix to obtain a global beamforming weight.

[0098] In an optional implementation, the learning optimization module 25 is specifically used for: Performing local loss function definition processing on the optimized precoding matrix to obtain interference power loss; Performing a stochastic gradient descent update process on a model parameter set in the local beam weight model according to the interference power loss to obtain a standardized beam weight model; The standardized beam weight model is subjected to a federal average aggregation process to obtain the global beamforming weight.

[0099] The iterative correction module 26 is used to perform phase error iterative correction processing on the global beamforming weight according to the detected radiation field distribution data to obtain an antenna array control signal.

[0100] In an optional implementation, the iterative correction module 26 is specifically used to: Performing acquisition signal preprocessing on the radiation field distribution data to obtain a radiation field vector; Performing antenna response matrix construction processing on the global beamforming weights to obtain an antenna array response matrix; Perform LM nonlinear iterative optimization processing according to the radiation field vector and the antenna array response matrix to obtain a phase error vector; A phase error compensation process is performed on the global beamforming weight according to the phase error vector to obtain the antenna array control signal.

[0101] It should be understood that the various variations and specific embodiments of the methods provided in the above embodiments are also applicable to the multi-band smart antenna array beamforming device of the present embodiment. Through the above detailed description of the multi-band smart antenna array beamforming method, those skilled in the art can clearly understand the implementation method of the multi-band smart antenna array beamforming device of the present embodiment. For the sake of brevity of the specification, it will not be described in detail here.

[0102] like Figure 3 , which is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application.

[0103] In a preferred embodiment of the present invention, the electronic device 3 may include, but is not limited to: a memory 31 , at least one processor 32 and at least one communication bus 33 .

[0104] Those skilled in the art should understand that Figure 3 The structure of the electronic device 3 shown does not constitute a limitation of the embodiment of the present invention, and the electronic device 3 may also include more or less other hardware or software than shown in the figure, or a different arrangement of components.

[0105] In some embodiments, the electronic device 3 is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application-specific integrated circuits, programmable gate arrays, digital processors and embedded devices.

[0106] It should be noted that the electronic device 3 is only an example, and other existing or future electronic products that are suitable for the present application should also be included in the protection scope of the present application and included here by reference.

[0107] In some embodiments, the memory 31 stores a computer program, and when the computer program is executed by the at least one processor 32, all or part of the steps in the multi-band smart antenna array beamforming method as described above are implemented. The memory 31 includes a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data. Further, the computer-readable storage medium may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function, and the like.

[0108] In some embodiments, the at least one processor 32 is the control core (ControlUnit) of the electronic device 3, and uses various interfaces and lines to connect various components of the entire electronic device 3, and executes various functions and processes data of the electronic device 3 by running or executing programs or modules stored in the memory 31, and calling data stored in the memory 31. For example, when the at least one processor 32 executes the computer program stored in the memory 31, it implements all or part of the steps of the multi-band smart antenna array beamforming method described in the embodiment of the present application; or implements all or part of the functions of the multi-band smart antenna array beamforming device. The at least one processor 32 can be composed of an integrated circuit, for example, it can be composed of a single packaged integrated circuit, or it can be composed of multiple integrated circuits with the same function or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips.

[0109] In some embodiments, the at least one communication bus 33 is configured to realize the connection and communication between the memory 31 and the at least one processor 32. Although not shown, the electronic device 3 may also include a power supply (such as a battery) for powering each component. Preferably, the power supply may be logically connected to the at least one processor 32 through a power management device, so as to realize the functions of managing charging, discharging, and power consumption management through the power management device. The power supply may also include any components such as one or more DC or AC power supplies, recharging devices, power failure detection circuits, power converters or inverters, power status indicators, etc. The electronic device 3 may also include a variety of sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be repeated here.

[0110] The above-mentioned integrated unit implemented in the form of a software function module can be stored in a computer-readable storage medium. The above-mentioned software function module is stored in a storage medium, including a number of instructions for enabling an electronic device (which can be a personal computer, electronic device, or network device, etc.) or a processor to execute part of the method described in each embodiment of the present application.

[0111] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only illustrative, for example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

[0112] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, and may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0113] The above are all preferred embodiments of the present application, and the protection scope of the present application is not limited thereto. Therefore, any equivalent changes made according to the structure, shape, and principle of the present application should be included in the protection scope of the present application.

Claims

1. A multi-band smart antenna array beamforming method, characterized in that: The method comprises: Perform dynamic frequency band switching processing according to the detected multi-dimensional environmental data to obtain frequency band decision parameters; Performing three-dimensional beam modeling processing according to the frequency band decision parameters to obtain a beam weight matrix; Performing hybrid precoding processing on the beam weight matrix according to a preset coding method to obtain a digital-analog precoding matrix; Performing temperature compensation correction processing on the digital-analog precoding matrix according to the detected array temperature distribution data to obtain an optimized precoding matrix; Performing federated learning optimization processing on a preset local beam weight model according to the optimized precoding matrix to obtain a global beamforming weight; The global beamforming weights are subjected to iterative phase error correction processing according to the detected radiation field distribution data to obtain an antenna array control signal.

2. The multi-band smart antenna array beamforming method according to claim 1, characterized in that: The performing dynamic frequency band switching processing according to the detected multi-dimensional environmental data to obtain the frequency band decision parameters includes: Perform parameter extraction and fusion processing on the detected multi-dimensional environmental data to obtain the environmental parameter matrix; The environment parameter matrix is ​​subjected to nonlinear mapping processing of a Gaussian kernel function by a preset support vector machine classification model to obtain a frequency band label and an initial frequency band decision parameter; Calculate the classification error based on the preset actual frequency band label and the frequency band label to obtain the classification error rate; According to a preset classification accuracy threshold and the classification error rate, an online learning update process is performed on the initial frequency band decision parameter to obtain the frequency band decision parameter.

3. The multi-band smart antenna array beamforming method according to claim 1, characterized in that: The performing three-dimensional beam modeling processing according to the frequency band decision parameters to obtain a beam weight matrix includes: Step S21, performing pilot signal transmission and reception processing according to the frequency band decision parameters by a preset channel estimator to obtain a multi-user channel state information matrix; Step S22, performing compressed sensing sparse reconstruction processing on the multi-user channel state information matrix to obtain an optimization objective function; Step S23, performing an iterative solution process of the optimization objective function using an alternating direction multiplier method to obtain an initial beam weight matrix; Step S24, performing Monte Carlo simulation processing on the initial beam weight matrix to obtain beam coverage data; Step S25, verifying the initial beam weight matrix according to a preset sidelobe level threshold and the beam coverage data, and updating the optimization objective function according to a preset method; Repeat step S23 to step S25 until the sidelobe level in the beam coverage data is less than the sidelobe level threshold, and use the initial beam weight matrix as the beam weight matrix.

4. The multi-band smart antenna array beamforming method according to claim 1, characterized in that: The preset coding method includes an alternating optimization algorithm, and the performing hybrid precoding processing on the beam weight matrix according to the preset coding method to obtain a digital-analog precoding matrix includes: Performing singular value decomposition processing on the beam weight matrix to obtain an initial digital precoding matrix and an initial analog precoding matrix; Performing alternating optimization iterative processing on the initial digital precoding matrix and the initial analog precoding matrix according to the alternating optimization algorithm to obtain a converged and optimized analog precoding matrix; The analog precoding matrix is ​​subjected to a phase shifter controlled codeword quantization process to obtain the digital-analog precoding matrix.

5. The multi-band smart antenna array beamforming method according to claim 1, characterized in that: The step of performing temperature compensation correction processing on the digital-analog precoding matrix according to the detected array temperature distribution data to obtain an optimized precoding matrix includes: A thermal expansion coefficient temperature difference model is constructed for the detected array temperature distribution data according to the preset LTCC process characteristic data to obtain a temperature-phase error model; Performing phase compensation matrix construction processing on the digital-analog precoding matrix by using the temperature-phase error model to obtain a diagonal matrix; A matrix product compensation process is performed on the digital-analog precoding matrix according to the diagonal matrix to obtain the optimized precoding matrix.

6. The multi-band smart antenna array beamforming method according to claim 1, characterized in that: The performing a federated learning optimization process on a preset local beam weight model according to the optimized precoding matrix to obtain a global beamforming weight includes: Performing local loss function definition processing on the optimized precoding matrix to obtain interference power loss; Performing a stochastic gradient descent update process on a model parameter set in a preset local beam weight model according to the interference power loss to obtain a standardized beam weight model; The standardized beam weight model is subjected to a federal average aggregation process to obtain the global beamforming weight.

7. The multi-band smart antenna array beamforming method according to claim 1, characterized in that: The performing phase error iterative correction processing on the global beamforming weight according to the detected radiation field distribution data to obtain the antenna array control signal includes: Performing acquisition signal preprocessing on the detected radiation field distribution data to obtain the radiation field vector; Performing antenna response matrix construction processing on the global beamforming weights to obtain an antenna array response matrix; Perform LM nonlinear iterative optimization processing according to the radiation field vector and the antenna array response matrix to obtain a phase error vector; A phase error compensation process is performed on the global beamforming weight according to the phase error vector to obtain the antenna array control signal.

8. A multi-band smart antenna array beamforming device, characterized in that: The device comprises: A decision parameter module, used for performing dynamic frequency band switching processing according to the detected multi-dimensional environmental data to obtain frequency band decision parameters; A beam weight module, used for performing three-dimensional beam modeling processing according to the frequency band decision parameters to obtain a beam weight matrix; A hybrid coding module, used for performing hybrid precoding processing on the beam weight matrix according to a preset coding method to obtain a digital-analog precoding matrix; A temperature correction module, used for performing temperature compensation correction processing on the digital-analog precoding matrix according to the detected array temperature distribution data to obtain an optimized precoding matrix; A learning optimization module, used for performing a federated learning optimization process on a preset local beam weight model according to the optimized precoding matrix to obtain a global beamforming weight; The iterative correction module is used to perform phase error iterative correction processing on the global beamforming weight according to the detected radiation field distribution data to obtain an antenna array control signal.

9. An electronic device, characterized in that: The electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the multi-band smart antenna array beamforming method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the multi-band smart antenna array beamforming method according to any one of claims 1 to 7 are implemented.

Images