Digital array radar multi-beam waveform fast generation method, device, equipment and medium

Abstract

The application relates to a digital array radar multi-beam waveform fast generation method, device, equipment and medium. The method comprises the following steps: constructing a multi-beam expected direction diagram based on array element parameters and expected beam parameters, constructing a teacher phase matrix, inputting the multi-beam expected direction diagram into a phase decoding network to obtain an array element phase matrix, calculating a teacher phase distillation loss according to the array element phase matrix and the teacher phase matrix, substituting the array element phase matrix into an array physical forward model to calculate a self-supervision physical loss, jointly training the network by using the two types of losses, inputting a new multi-beam expected direction diagram to be processed into the trained network to obtain a transmission phase matrix and generate a constant envelope transmission waveform. By adopting the method, the multi-beam waveform can be quickly generated under the premise of ensuring the direction diagram matching and low correlation performance, and the online waveform agility capability of the digital array radar is improved.

Description

Technical Field

[0001] This application relates to the field of digital array radar waveform design technology, and in particular to a method, apparatus, device and medium for rapid generation of multi-beam waveforms for digital array radar. Background Technology

[0002] Digital array radars transmit coherent or orthogonal waveforms through multiple array elements and form a desired transmission pattern in space using the phase, amplitude, and time coding relationships between the elements. To meet the needs of different target search, tracking, anti-jamming, and counter-reconnaissance missions, the radar transmission pattern often needs to change rapidly according to the environment and mission requirements. Traditional waveform design methods typically involve constructing a non-convex optimization objective function and iteratively solving for the element waveforms or covariance matrix, then recovering the specific waveform from the covariance matrix. This process is computationally intensive and may introduce errors.

[0003] With the development of deep learning, directly learning the inverse mapping from the desired radiation pattern to the array element waveform using neural networks has become a feasible approach. However, digital array radar waveform design is not a single-valued inverse problem; the same radiation pattern usually corresponds to multiple sets of phase solutions. If the network only learns radiation pattern fitting, it is easy to obtain phases with usable radiation patterns but high autocorrelation or cross-correlation sidelobes; if it only learns a certain set of teacher phases, it is easy to have insufficient generalization when the number of beams and beam angles change. Therefore, a training method is needed that can both enable the network to learn stable phase generation rules and utilize array physical loss to constrain the performance of the final output waveform.

[0004] Furthermore, the method of generating the desired radiation pattern has a decisive impact on network training. If the desired radiation pattern directly uses a truncated rectangular main lobe or artificially sets the side lobe floor, it may be inconsistent with the actual physical model of the array, leading to instability in the network's learning objectives. Summary of the Invention

[0005] Therefore, it is necessary to provide a method, apparatus, device, and medium for rapid generation of multi-beam waveforms for digital array radar that can take into account both pattern matching and low correlation constraints, support flexible configuration of multi-beam power, and have online rapid inference capabilities, in order to address the above-mentioned technical problems.

[0006] A method for rapid generation of multi-beam waveforms in a digital array radar, the method comprising: Obtain the array element parameters of the digital array radar, as well as the desired beam parameters during training; Based on the array element parameters and the desired beam parameters, a corresponding multi-beam desired radiation pattern is constructed; Based on the array element parameters and desired beam parameters, each beam is sorted by center angle and the number of snapshots is allocated according to power weight. A phase template corresponding to the beam center is constructed, and the phase offset of each snapshot in the group is generated by indexing the group number using a fixed codebook. After superimposing the phase template and the phase offset, the teacher phase matrix is ​​obtained after physical constraint optimization with the desired multi-beam radiation pattern as the optimization target. The multi-beam desired radiation pattern is input into the phase decoding network. In the phase decoding network, the encoder first extracts global features, and then the group query decoder predicts the group phase template, the snapshot offset within the group, the group size and the group activation state of each beam group, and then obtains the element phase matrix. The teacher phase distillation loss is calculated based on the array element phase matrix and the teacher phase matrix. The self-supervised physical loss, including the pattern loss and the related sidelobe loss, is calculated after substituting the array element phase matrix into the array physical forward model. The phase decoding network is jointly trained using the self-supervised physical loss and the teacher phase distillation loss to obtain the trained phase decoding network. The new multi-beam desired radiation pattern to be processed is input into the trained phase decoding network to obtain the transmit phase matrix of each snapshot of each element on the digital array radar, and a constant envelope transmit waveform is generated based on the transmit phase matrix.

[0007] In one embodiment, constructing the corresponding multi-beam desired radiation pattern based on the array element parameters and the desired beam parameters includes: For each desired beam, the physically realizable main lobe region of the desired beam is determined based on the number of array elements and the spacing between array elements in the array element parameters. Only the power value within the main lobe region is retained, and the desired power outside the main lobe region is set to zero. The main lobe power of each beam is synthesized and normalized according to the power weight in the desired beam parameters to obtain the desired multi-beam radiation pattern.

[0008] In one embodiment, the physically achievable main lobe region is the first null main lobe region corresponding to the beam center, and the boundary of the first null main lobe region is determined by the product of the number of array elements and the array element spacing. If the main lobe regions of multiple beams overlap, the maximum value of the normalized weighted power of each beam is taken as the expected power at the corresponding angle.

[0009] In one embodiment, the input to the phase decoding network further includes a sequence of decibel values ​​and angular Fourier features of the multi-beam desired pattern; The encoder is a Transformer encoder. The phase decoding network extracts the beam center, peak value, width, and activation mask from the multi-beam desired radiation pattern as a radiation pattern descriptor, and fuses the radiation pattern descriptor with group query features; The group query decoder obtains the context information of each beam group from the global features output by the encoder through a cross-attention mechanism, and predicts the parameters of each beam group based on the context information and the radiation pattern descriptor.

[0010] In one embodiment, the teacher phase distillation loss includes group phase template loss, group snapshot offset loss, group size loss, group activation loss, and absolute phase loss, which are used to guide the output of the phase decoding network to approximate the generation rules of the teacher phase matrix.

[0011] In one embodiment, the pattern loss includes pattern main lobe shape loss and main lobe spurious peak suppression loss. The pattern main lobe shape loss is used to constrain the actual transmit pattern corresponding to the network output phase within the main lobe region to match the desired multi-beam direction. Figure 1 Therefore, the main lobe spurious peak suppression loss is used to suppress unwanted radiation outside the main lobe region; The correlation sidelobe loss includes autocorrelation peak sidelobe loss and cross-correlation peak sidelobe loss. The autocorrelation peak sidelobe loss is used to constrain the autocorrelation sidelobe level of the same array element waveform under different delays, and the cross-correlation peak sidelobe loss is used to constrain the cross-correlation sidelobe level between different array element waveforms.

[0012] In one embodiment, the joint training uses fixed-beam-number anchor samples and random multi-beam samples to participate in the training. The fixed beam number anchor point sample is: according to the number of beams in the desired beam parameters, take each fixed integer value from 1 to the preset maximum number of beam groups, and set the center angle and power weight of each beam for each beam number to generate the corresponding multi-beam desired radiation pattern and teacher phase matrix as a training sample pair; The random multi-beam sample is generated by randomly selecting values ​​within their respective ranges according to the number of beams, the center angle of each beam, and the power weight of each beam in the desired beam parameters, and using them as a training sample pair. The joint training includes a fixed-beam-number anchor training phase and a random multi-beam generalization training phase. In the fixed-beam-number anchor training phase, corresponding multi-beam expected radiation patterns and teacher phase matrices are generated for fixed-beam-number anchor samples, and the phase decoding network is trained using teacher phase distillation loss and self-supervised physical loss. In the random multi-beam generalization training phase, the number of beams, beam center angles, and power weights are randomly generated to construct random multi-beam expected radiation patterns, and the phase decoding network is fine-tuned using the self-supervised physical loss. At the same time, fixed-beam-number anchor samples are added to suppress the network's forgetting of existing beam configurations.

[0013] This application also provides a digital array radar multi-beam waveform rapid generation device, the device comprising: The training parameter acquisition module is used to acquire the array element parameters of the digital array radar, as well as the desired beam parameters during training. A multi-beam desired pattern generation module is used to construct a corresponding multi-beam desired pattern based on the array element parameters and desired beam parameters. The teacher phase matrix generation module is used to sort each beam by center angle and allocate the number of snapshots according to power weight based on the array element parameters and desired beam parameters, construct a phase template with spatial guidance corresponding to the beam center, and generate the phase offset of each snapshot in the group by group number index using a fixed codebook. After superimposing the phase template and the phase offset, the teacher phase matrix is ​​obtained after physical constraint optimization with the desired multi-beam radiation pattern as the optimization target. The network output module is used to input the multi-beam desired radiation pattern into the phase decoding network. In the phase decoding network, the encoder first extracts global features, and then the group query decoder predicts the group phase template, the snapshot offset within the group, the group size and the group activation state of each beam group, and then obtains the array element phase matrix. The network training module is used to calculate the teacher phase distillation loss based on the array element phase matrix and the teacher phase matrix, and to calculate the self-supervised physical loss including the pattern loss and the related sidelobe loss after substituting the array element phase matrix into the array physical forward model. The self-supervised physical loss and the teacher phase distillation loss are used to jointly train the phase decoding network to obtain the trained phase decoding network. The multi-beam waveform rapid generation module is used to input the new multi-beam desired radiation pattern to be processed into the trained phase decoding network to obtain the transmit phase matrix of each snapshot of each element on the digital array radar, and generate a constant envelope transmit waveform based on the transmit phase matrix.

[0014] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps in the above-described method for rapid generation of multi-beam waveforms for digital array radar.

[0015] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in a method for rapid generation of multi-beam waveforms for digital array radar.

[0016] The aforementioned method, apparatus, device, and medium for rapid generation of multi-beam waveforms in digital array radar, by acquiring array element parameters and desired beam parameters and constructing a multi-beam desired radiation pattern that conforms to the physical response characteristics of the array, provides a stable and physically reliable training target for the network, avoiding training bias caused by non-physical truncation of the desired radiation pattern. Furthermore, by constructing a teacher phase matrix that fuses spatial steering phase and a fixed codebook and is optimized by physical constraints, it provides stable and learnable generation rules for the multi-inverse problem, effectively alleviating the network's generalization deficiency when the multi-beam configuration changes. The desired radiation pattern is then input into the phase decoding network and trained using a combination of teacher phase distillation loss and a self-supervised physical loss including radiation pattern loss and related sidelobe loss. This allows the network to learn stable teacher rules while its output phase is directly constrained by the array's physical forward model, ensuring that the final waveform simultaneously meets the requirements of radiation pattern matching and low correlation performance. After training, only a single forward propagation of the new desired radiation pattern is needed to output the array element phase matrix and generate a constant envelope waveform, thus fundamentally overcoming the shortcomings of traditional iterative optimization, which is too time-consuming and unable to meet the real-time requirements of online waveform agility. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating a method for rapid generation of multi-beam waveforms for digital array radar in one embodiment. Figure 2 This is a schematic diagram of the process framework of the method in one embodiment; Figure 3 A schematic diagram of generating the desired pattern of the first null point of the main lobe in one embodiment; Figure 4 This is a schematic diagram of the phase decoding network in one embodiment; Figure 5 This is a schematic diagram of teacher distillation and self-supervised physical loss training in one embodiment; Figure 6 This is a structural block diagram of a digital array radar multibeam waveform rapid generation device in one embodiment; Figure 7 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0019] To address the problems in existing technologies, such as the high computational cost and difficulty in meeting the real-time requirements of waveform agility in traditional iterative optimization methods, and the instability and poor generalization of deep learning methods due to multiple solutions to the inverse radiation pattern problem, as well as the impact on training performance caused by inconsistencies between the desired radiation pattern generation method and the array physical model, this application addresses these issues. Figure 1 As shown, a method for rapid generation of multi-beam waveforms in a digital array radar is provided, which specifically includes the following steps: Step S100: Obtain the array element parameters of the digital array radar, as well as the desired beam parameters during training.

[0020] Step S110: Based on the array element parameters and the desired beam parameters, construct the corresponding multi-beam desired radiation pattern.

[0021] Step S120: Based on the array element parameters and the desired beam parameters, sort each beam by center angle and allocate the number of snapshots according to power weight. Construct a phase template for spatial guidance corresponding to the beam center. Use a fixed codebook to generate the phase offset of each snapshot in the group according to the group number index. After superimposing the phase template and the phase offset, take the desired multi-beam radiation pattern as the optimization target and optimize it through physical constraints to obtain the teacher phase matrix.

[0022] In step S130, the multi-beam desired radiation pattern is input into the phase decoding network. In the phase decoding network, the encoder first extracts global features, and then the group query decoder predicts the group phase template, the snapshot offset within the group, the group size and the group activation state of each beam group, thereby obtaining the element phase matrix.

[0023] Step S140: Calculate the teacher phase distillation loss based on the array element phase matrix and the teacher phase matrix. Substitute the array element phase matrix into the array physical forward model to calculate the self-supervised physical loss, which includes pattern loss and related sidelobe loss. Use the self-supervised physical loss and the teacher phase distillation loss to jointly train the phase decoding network and obtain the trained phase decoding network.

[0024] Step S150: Input the new multi-beam desired radiation pattern to be processed into the trained phase decoding network to obtain the transmit phase matrix of each element and each snapshot on the digital array radar, and generate a constant envelope transmit waveform based on the transmit phase matrix.

[0025] In this method, to address the problem of unstable network training and poor generalization caused by multiple solutions to the inverse pattern problem, a teacher phase matrix is ​​constructed as a stable and learnable generation rule. The output of the phase decoding network is jointly constrained by the teacher phase distillation loss and the array physical forward model self-supervised loss, enabling the network to achieve fast inference while maintaining good pattern matching and low-correlation performance. The overall process of this method can also be found in [reference needed]. Figure 2 .

[0026] In step S100, the array element parameters include the number of array elements of the digital array radar. and array element spacing The desired beam parameters during training include the center angles of each desired beam. and the corresponding power weights ,in Indicates the beam number. This represents the total number of beams in the current training samples. These parameters collectively form the basic inputs for subsequent desired pattern generation and teacher phase construction.

[0027] In step S110, a corresponding multi-beam desired radiation pattern is constructed based on the array element parameters and the desired beam parameters.

[0028] Specifically, let the digital array radar be... Uniform linear array with element spacing of [missing information] The angle sampling range is 90° to 90°. A single-beam main lobe is generated using the sinc-type array response corresponding to the full element aperture. For each desired beam, the number of elements in the element parameters is used... and array element spacing The physically realizable main lobe region of the beam is determined, specifically the first null main lobe region corresponding to the beam center. The boundary of the first null main lobe region is determined by the number of array elements. and array element spacing The product determines that the first null support region of the beam satisfies:

[0029] In the above formula, The physical main lobe power of the beam is preserved within the internal structure. The desired power is set to zero. For multiple desired beams, the main lobe power of each beam is synthesized according to power weights, and the larger value of each beam's contribution is taken at each angle sampling point. After normalization, the multi-beam desired radiation pattern is obtained. If the main lobe regions of multiple beams overlap, the maximum value of the normalized weighted power of each beam is taken as the desired power at that angle. The multi-beam desired radiation pattern is then expressed as:

[0030] In one embodiment, taking a 32-element array as an example, the desired radiation pattern is as follows: Figure 3 As shown, its horizontal axis range is 90° to 90°, vertical axis according to Display from 60dB to 0dB, main lobe displayed to The expected power data outside the main lobe is strictly set to zero at 60dB, and no non-zero sidelobe power floor is set. To facilitate logarithmic domain calculations and graphical displays, a preset lower limit is used when converting power to decibel values; the lower limit is only used to avoid taking the logarithm of zero power and to limit the display range, and does not indicate the existence of non-zero expected sidelobe power outside the main lobe.

[0031] In step S120, let the total number of snapshots be... First, all beams are sorted according to their center angles. After sorting, the power weights of each beam are determined. Total number of snapshots Beams are assigned to different beam groups, and the number of snapshots in each group is determined. Beam groups with higher power weights are assigned more snapshots. For the [number]th [beam group]... A beam group is constructed with an angle relative to the center of the beam. The corresponding space-guided phase template is represented as:

[0032] In the above formula, For array element sequence number, The element spacing is normalized. Simultaneously, a fixed low-correlation codebook is used to generate the phase offset of each snapshot within a group, indexed by the sorted group number. This fixed low-correlation codebook is a pre-defined stable codebook; different group numbers correspond to different codebook entries, obtained by indexing the group number, rather than being temporarily generated by random perturbations. The spatially guided phase template is superimposed with the snapshot phase offsets within the group to obtain the initial phase matrix for each group.

[0033] Furthermore, using the desired multi-beam radiation pattern as the optimization objective, the initial phase matrices of each group are substituted into the array physical forward model to calculate the corresponding transmission radiation pattern. This pattern is then optimized through radiation pattern fitting, suppression of unwanted radiation outside the main lobe, and constraints on autocorrelation peak sidelobes. The fixed low-correlation codebook is pre-screened based on autocorrelation and cross-correlation performance, ensuring that the teacher's initial phase has a stable low-correlation coding structure. The cross-correlation peak sidelobes are further constrained during the subsequent self-supervised physical training of the phase decoding network. The optimization objective is to make the main lobe region of the actual transmission radiation pattern approximate the desired radiation pattern while satisfying the preset autocorrelation and cross-correlation sidelobe constraints. After optimization, the teacher phase matrix is ​​obtained. This teacher phase matrix satisfies the main lobe shape requirements of the desired radiation pattern and possesses a stable, low-correlation coding structure, providing learnable generation rules for the phase decoding network.

[0034] In one embodiment, the digital array radar employs a 32-element uniform linear array, with each group of transmitted waveforms containing 256 snapshots. For a single beam group, its teacher phase matrix has a dimension of 32×256. For training samples containing multiple beam groups, the 256 snapshots are distributed among the groups according to the power weights of each beam group, with the sum of the number of snapshots in each group being 256. Subsequently, the phase sub-matrices of each group are concatenated along the snapshot dimension according to the sorting result of the beam center angle to form a 32×256 array. The complete teacher phase matrix of 256.

[0035] In step S130, as Figure 4 As shown, the phase decoding network includes an angle feature encoding layer, an encoder, a group query decoder, a group phase template output head, a group snapshot offset output head, a group size output head, and a group activation output head. The encoder is a Transformer encoder. The network input is the multi-beam desired radiation pattern constructed in step S110.

[0036] In this embodiment, the phase decoding network extracts the beam center, peak value, width, and activation mask from the multi-beam desired radiation pattern as a radiation pattern descriptor, and fuses the radiation pattern descriptor with the group query features. The group query decoder obtains the context information of each beam group from the global features output by the encoder through a cross-attention mechanism, and predicts the parameters of each beam group based on the context information and the radiation pattern descriptor.

[0037] In this embodiment, for angle sampling points The network input features include the power values ​​of the desired radiation pattern. Desired radiation pattern decibel value The input to the angle feature encoding layer consists of several order angular Fourier features, which together form the input of the angle feature encoding layer.

[0038] Furthermore, the angle feature encoding layer concatenates and encodes the above three types of features before inputting them into the Transformer encoder. The Transformer encoder extracts global context features of the radiation pattern across the entire angular range using a self-attention mechanism, capturing the spatial relationships and power distribution features between beams. Subsequently, the group query decoder receives the global features output by the encoder and combines them with the beam center, peak value, width, and activation mask extracted from the desired radiation pattern as descriptors, using a cross-attention mechanism to predict the parameters of each beam group.

[0039] Furthermore, the group query decoder is configured with a preset maximum number of beam groups. In one embodiment, the preset maximum number of beam groups is set to 5 to cover single-beam to multi-beam configurations. For each beam group, the group phase template output head predicts the M-dimensional group phase template of the group, the group snap offset output head predicts the offset of each snap within the group relative to the group's starting position, the group size output head predicts the number of snaps in the group, and the group activation output head predicts whether the group is activated (i.e., whether a corresponding desired beam exists). Finally, based on the predicted group size, the prediction results of each group are concatenated and expanded along the snap dimension to form a complete M×L element phase matrix.

[0040] In the above formula, For the first The phase sub-matrix of each beam group has a dimension of , This represents the predicted number of snapshots for this group, and The phase sub-matrix of each beam group is obtained by superimposing its group phase template with the offsets of each snapshot within the group. Based on the element phase matrix, further... The constant envelope emission waveform is obtained.

[0041] In step S140, the array element phase matrix output by the network in step S130 is sent to two parallel branches: the teacher phase distillation branch and the self-supervised physical loss branch.

[0042] In this embodiment, the training process of this method uses teacher phase distillation loss to provide stable learnable generative rules for the network, thereby alleviating the training instability caused by multiple solutions to the inverse pattern problem. Simultaneously, self-supervised physical loss ensures that the network output waveform is directly constrained by the array physical forward model, guaranteeing that the final waveform meets radar engineering requirements in terms of pattern matching and related sidelobe performance. These two types of losses are jointly optimized during network training, forming a dual-track learning paradigm of "rule-guided + physical verification." The following combines... Figure 5 The training process is explained in detail.

[0043] In this embodiment, the teacher phase distillation loss is used to guide the output of the phase decoding network to approximate the generation rule of the teacher phase matrix. The teacher phase distillation loss includes group phase template loss. Group snap offset loss Group size loss Group activation loss and absolute phase loss Among them, group phase template loss The phase templates for each beam group used to constrain network prediction are consistent with the phase templates of the corresponding groups in the teacher's phase matrix; group snapshot offset loss The snapshot offsets within each group used to constrain network predictions are consistent with the snapshot offsets of the corresponding groups in the teacher phase matrix; group size loss. The number of snapshots in each group used to constrain network prediction is consistent with the number of snapshots in the corresponding group in the teacher's phase matrix; group activation loss. The activation states of each beam group used to constrain network prediction are consistent with the activation states of the corresponding groups in the teacher phase matrix; absolute phase loss. The overall phase matrix used to constrain the network output is close to the teacher's phase matrix in terms of absolute phase distance, thus maintaining the continuity of the overall phase.

[0044] Notation Conventions: For ease of reading, the notation used in the formulas below will be consistent. Let the number of matrix elements be . The number of quick shots is The complete phase matrix output by the network is:

[0045] The corresponding constant envelope waveform is:

[0046] Because phase has Due to periodicity, this paper does not directly compare phases using ordinary squared errors, but instead uses circular phase errors:

[0047] When two phases differ When it is an integer multiple of, This format is more suitable for phase code learning.

[0048] In this embodiment, the role of teacher phase distillation is to enable the neural network to learn the "structured rules" used by the teacher optimizer when generating the phase matrix. These rules include spatial phase templates, intra-group time offsets, group snapshot allocation, group activation states, and complete phase encoding.

[0049] Let the set of effective beam groups provided by the teacher be... The maximum number of groups is . No. The teacher space template, teacher time offset, and teacher snapshot count for each valid group are respectively denoted as:

[0050] The network corresponding prediction is:

[0051] In the above formula, For group size logit, The group is activated by logit. The complete teacher phase matrix is:

[0052] In this embodiment, the total loss of the teacher phase distillation is:

[0053] Specifically, the spatially guided phase template predicted by the group phase template loss constrained network is close to the teacher template:

[0054] To put it simply, Ensure that the spatial pointing of each beam group does not deviate.

[0055] Specifically, the temporal phase shift predicted by the group snapshot offset loss constrained network is close to the teacher's phase shift:

[0056] To put it simply, The main focus is on studying timecode structures to reduce autocorrelation and cross-correlation sidelobes.

[0057] Specifically, the group size loss constrains the network's prediction of the proportion of snapshots in each group to approximate the teacher allocation. Let:

[0058] but:

[0059] In the above formula, This represents the set of invalid groups. The first term ensures the correct proportion of snapshots, and the second term suppresses invalid groups from occupying snapshots. For invalid beamgroups, the definition is... ;when When the set is empty, the second term is zero.

[0060] Specifically, the group activation loss uses a binary cross-entropy constrained network to determine which groups should be activated. Let... If the effective group indicator variable for teachers is:

[0061] To put it simply, How many beam groups are needed for the network to learn the current target pattern?

[0062] Specifically, the absolute phase loss is expressed as:

[0063] This approach does not only consider the radiation pattern but also directly maintains the overall phase code structure, which helps to maintain phase continuity and related sidelobe performance.

[0064] Specifically, the self-supervised physical loss does not depend on the teacher phase. Instead, it substitutes the phase matrix output by the network into the array physical forward model to directly calculate the RF performance indicators of the actual transmitted waveform. This includes pattern loss and related sidelobe loss, where pattern loss includes pattern main lobe shape (i.e., pattern fitting) loss. and the loss of suppression of false peaks outside the main lobe The main lobe shape loss is used to constrain the actual transmit pattern corresponding to the network output phase within the main lobe region, relative to the desired multi-beam direction. Figure 1 The main lobe spurious peak suppression loss is used to suppress unwanted radiation outside the main lobe region. The correlated sidelobe loss includes autocorrelation peak sidelobe loss. and cross-correlation peak sidelobe loss These are used to constrain the autocorrelation sidelobe levels of the same array element waveform under different delays, and the cross-correlation sidelobe levels between different array element waveforms, respectively.

[0065] Specifically, the self-supervised physical loss no longer directly mimics the teacher phase, but instead feeds the phase matrix output by the network into the array physical forward model to directly check whether the actual radiation pattern and related sidelobes meet the requirements.

[0066] Let the network output pattern be The desired direction pattern is Then convert both to the dB domain:

[0067] The current target pattern of this project only retains the first null region of the main lobe, and the target value outside the main lobe is zero. Therefore, the following definition is made:

[0068] In the above formula, It is the desired main lobe region. It is the undesired radiation region outside the main lobe.

[0069] The pattern loss consists of three parts: main lobe shape matching, main lobe peak ranking, and main lobe spurious peak suppression.

[0070] First, the weighted dB-domain shape error of the main lobe region is:

[0071] In the above formula, It is the number of angle sampling points. Used to increase the weight of the main lobe region.

[0072] set up Main lobe support area The set of connected main lobes obtained from the decomposition. If the lower limit of the main lobe peak value is taken, then the main lobe peak value ranking loss is:

[0073] In the above formula, the first term requires that the peak value of the main lobe be higher than the maximum false peak outside the main lobe, and the second term requires that the peak value of the main lobe be no lower than a specified lower limit.

[0074] Therefore, the overall loss of the directional pattern is:

[0075] In the above formula, It is the upper limit of the false peak outside the main lobe.

[0076] Explicit spurious peak suppression loss: Specifically, to further suppress local peaks outside the main lobe, an additional top-K spurious peak loss is used:

[0077] In the above formula, This represents the K angle points in the outer region of the main lobe that have the largest over-limit penalty. This is an explicit pseudo-peak threshold.

[0078] Specifically, the autocorrelation sidelobe loss is used to constrain the autocorrelation sidelobes of the same array element waveform under different delays:

[0079] In the above formula, It is the autocorrelation sidelobe training threshold. For example, it can be -16dB during training and -15dB for threshold checking during evaluation.

[0080] Specifically, the cross-correlation sidelobe loss is used to constrain the cross-correlation sidelobes of different array element waveforms:

[0081] In the above formula, It is the cross-correlation sidelobe training threshold.

[0082] Specifically, the total loss for physical fine-tuning, considering pattern fitting, spurious peak suppression, autocorrelation sidelobes, cross-correlation sidelobes, and reference phase canonicalization, is as follows:

[0083] In the above formula, This is the output of the model before fine-tuning, frozen. The reference phase regularization term is used to prevent random generalization fine-tuning from breaking existing low-correlation coding rules.

[0084] The calculation of physical loss is based on the array physical forward model, which is defined as follows: Let the first The waveform of each element in a quick snapshot is as follows:

[0085] And direction The steering vector at that location is ,

[0086] The normalized emission pattern is then:

[0087] The autocorrelation peak sidelobes are calculated from the correlation results of the same array element waveform under different delays, while the cross-correlation peak sidelobes are calculated from the correlation results of different array element waveforms under different delays.

[0088] Furthermore, the overall loss function can be expressed as a weighted sum of the loss terms:

[0089]

[0090] In the above formula, ~ The weighting coefficients for each term in the teacher's phase distillation loss are... ~ The weighting coefficients for each item of self-supervised physical loss are preset according to engineering requirements.

[0091] In this embodiment, the joint training adopts a strategy of mixing fixed-beam anchor samples with random multi-beam samples to balance the stability of the network under the basic beam configuration and the generalization ability under the unknown beam configuration.

[0092] Specifically, the joint training includes a fixed-beam-number anchor training phase and a randomized multi-beam generalization training phase. In the fixed-beam-number anchor training phase, corresponding multi-beam expected radiation patterns and teacher phase matrices are generated for the fixed-beam-number anchor samples, and the phase decoding network is trained using teacher phase distillation loss and self-supervised physical loss. In the randomized multi-beam generalization training phase, the number of beams, beam center angles, and power weights are randomly generated to construct randomized multi-beam expected radiation patterns, and the phase decoding network is fine-tuned using self-supervised physical loss. Simultaneously, fixed-beam-number anchor samples are added to suppress the network's forgetting of existing beam configurations.

[0093] Specifically, fixed-beam-number anchor point samples are generated as follows: Based on the number of beams in the desired beam parameters, each fixed integer value from 1 to the preset maximum number of beam groups is selected, such as 1, 2, 3, 4, 5. For each beam number, the center angle and power weight of each beam are fixedly set (e.g., single beam at 0°, dual beam at ±20°, triple beam at -30°, 0°, 30°, etc., typical configurations). This generates the corresponding multi-beam desired radiation pattern and teacher phase matrix as a training sample pair. Since the number of beams, center angle, and power weights corresponding to this type of sample are all fixed and typical configurations, the network continuously receives supervision from these basic samples during training. This prevents performance degradation under simple configurations due to interference from a large number of random samples, thus maintaining stable output under each basic beam number configuration.

[0094] Specifically, random multi-beam samples are generated as follows: based on the number of beams, the center angle of each beam, and the power weight of each beam in the desired beam parameters, values ​​are randomly selected within their respective ranges (e.g., the number of beams is randomly selected from 1 to 5, the center angle of each beam is randomly selected within a preset effective angle range, and the power weight of each beam is randomly generated under normalization constraints). This generates a corresponding multi-beam desired radiation pattern and a teacher phase matrix as a training sample pair. Since this type of sample covers a large number of angle and power combinations for each number of beams, the network can encounter a rich variety of beam configurations during training, truly learning the underlying mapping rules of inferring the element phase from the shape of the desired radiation pattern, thereby improving the ability of the generated phase to fit the desired radiation pattern under new beam configuration conditions.

[0095] In this embodiment, during the fixed-beam-number anchor training phase, the teacher phase distillation loss and self-supervised physical loss are jointly calculated for anchor samples with different fixed-beam-numbers at each parameter update. During the random multi-beam generalization training phase, each training iteration includes random multi-beam sample batches and fixed-beam-number anchor sample batches, and physical fine-tuning is primarily performed using pattern loss, main lobe spurious peak suppression loss, autocorrelation peak sidelobe loss, cross-correlation peak sidelobe loss, and a reference phase regularization term. The network learns stable generation rules for the teacher phase through teacher phase distillation loss, and through self-supervised physical loss, the pattern matching and correlation performance of the output waveform are directly constrained by the array physical model. The joint optimization of these two types of losses guides the network to converge under the dual supervision of "rule guidance" and "physical verification." After training, a trained phase decoding network is obtained.

[0096] In this embodiment, the training angle grid is taken as follows: 361 sampling points within 90° to 90°, or 12001 or more sampling points for high-resolution plotting or acceptance grids. The maximum number of network beamgroups is set to 5 to cover single-beam to multi-beam configurations.

[0097] In the aforementioned method for rapid generation of multi-beam waveforms for digital array radar, the array element parameters and desired beam parameters are obtained in step S100. Combined with the multi-beam desired radiation pattern constructed based on the array physical response in step S110 (only the first null main lobe region is retained, and the main lobe external power is set to zero), this avoids the training target unreachability problem caused by manually setting non-physical truncation of the main lobe or side lobe floor in traditional methods, ensuring the network training target is consistent with the array physical model. In step S120, a teacher phase matrix is ​​constructed by fusing a spatially guided phase template and a fixed low-correlation codebook and optimizing it through physical constraints. This is then combined with the teacher phase distillation loss in step S140 to supervise the network, enabling it to learn a stable, interpretable, and consistent phase generation method. This effectively overcomes the network training instability and insufficient generalization caused by multiple solutions to the inverse radiation pattern problem. Furthermore, in step S140, the network simultaneously introduces… By incorporating self-supervised physical loss, the network output phase is substituted into the array physical forward model to calculate the pattern loss and related sidelobe loss, and then jointly optimized. This ensures that the array element phase matrix output by the network not only approximates the generation rules of the teacher phase, but also that the corresponding actual transmitted waveform is directly constrained by the array physical model in terms of pattern matching, main lobe external spurious radiation suppression, autocorrelation sidelobe level, and cross-correlation sidelobe level, thereby ensuring that the RF performance of the final waveform meets the radar engineering requirements. Finally, in step S150, the trained phase decoding network is used for inference. Only a single forward propagation of the new desired pattern to be processed is needed to directly obtain the transmitted phase matrix of each element and each snapshot, and generate a constant envelope waveform. This avoids the huge computational overhead caused by repeatedly solving non-convex problems in traditional iterative optimization, significantly improving the multi-beam waveform generation speed and enabling digital array radar to quickly complete waveform agility during mission switching. Therefore, this method achieves fast, stable, and highly generalizable end-to-end generation from the desired pattern to the array element phase matrix while taking into account pattern matching and low correlation performance.

[0098] It should be understood that, although Figure 1 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 1 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0099] In one embodiment, such as Figure 6 As shown, a digital array radar multi-beam waveform rapid generation device is provided, comprising: a training parameter acquisition module 200, a multi-beam desired pattern generation module 210, a teacher phase matrix generation module 220, a network output module 230, a network training module 240, and a multi-beam waveform rapid generation module 250, wherein: The training parameter acquisition module 200 is used to acquire the array element parameters of the digital array radar, as well as the desired beam parameters during training. The multi-beam desired pattern generation module 210 is used to construct a corresponding multi-beam desired pattern based on the array element parameters and desired beam parameters. The teacher phase matrix generation module 220 is used to sort each beam by center angle and allocate the number of snapshots according to power weight based on the array element parameters and desired beam parameters, construct a phase template with spatial guidance corresponding to the beam center, and generate the phase offset of each snapshot in the group by group number index using a fixed codebook. After superimposing the phase template and the phase offset, the teacher phase matrix is ​​obtained after physical constraint optimization with the desired multi-beam radiation pattern as the optimization target. The network output module 230 is used to input the multi-beam desired radiation pattern into the phase decoding network. In the phase decoding network, the encoder first extracts global features, and then the group query decoder predicts the group phase template, the snapshot offset within the group, the group size and the group activation state of each beam group, and then obtains the array element phase matrix. The network training module 240 is used to calculate the teacher phase distillation loss based on the array element phase matrix and the teacher phase matrix, calculate the self-supervised physical loss including the pattern loss and the related sidelobe loss after substituting the array element phase matrix into the array physical forward model, and jointly train the phase decoding network using the self-supervised physical loss and the teacher phase distillation loss to obtain the trained phase decoding network. The multi-beam waveform rapid generation module 250 is used to input the new multi-beam desired radiation pattern to be processed into the trained phase decoding network to obtain the transmit phase matrix of each snapshot of each element on the digital array radar, and generate a constant envelope transmit waveform based on the transmit phase matrix.

[0100] Specific limitations regarding the rapid generation device for multi-beam waveforms of digital array radar can be found in the limitations of the rapid generation method for multi-beam waveforms of digital array radar described above, and will not be repeated here. Each module in the aforementioned rapid generation device for multi-beam waveforms of digital array radar can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in hardware or independently of the processor in a computer device, or stored in software in the memory of a computer device, so that the processor can call and execute the operations corresponding to each module.

[0101] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 7 As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for rapid generation of multi-beam waveforms for digital array radar. The display screen can be a liquid crystal display (LCD) or an e-ink display. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0102] Those skilled in the art will understand that Figure 7 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0103] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps: Obtain the array element parameters of the digital array radar, as well as the desired beam parameters during training; Based on the array element parameters and the desired beam parameters, a corresponding multi-beam desired radiation pattern is constructed; Based on the array element parameters and desired beam parameters, each beam is sorted by center angle and the number of snapshots is allocated according to power weight. A phase template corresponding to the beam center is constructed, and the phase offset of each snapshot in the group is generated by indexing the group number using a fixed codebook. After superimposing the phase template and the phase offset, the teacher phase matrix is ​​obtained after physical constraint optimization with the desired multi-beam radiation pattern as the optimization target. The multi-beam desired radiation pattern is input into the phase decoding network. In the phase decoding network, the encoder first extracts global features, and then the group query decoder predicts the group phase template, the snapshot offset within the group, the group size and the group activation state of each beam group, and then obtains the element phase matrix. The teacher phase distillation loss is calculated based on the array element phase matrix and the teacher phase matrix. The self-supervised physical loss, including the pattern loss and the related sidelobe loss, is calculated after substituting the array element phase matrix into the array physical forward model. The phase decoding network is jointly trained using the self-supervised physical loss and the teacher phase distillation loss to obtain the trained phase decoding network. The new multi-beam desired radiation pattern to be processed is input into the trained phase decoding network to obtain the transmit phase matrix of each snapshot of each element on the digital array radar, and a constant envelope transmit waveform is generated based on the transmit phase matrix.

[0104] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor: Obtain the array element parameters of the digital array radar, as well as the desired beam parameters during training; Based on the array element parameters and the desired beam parameters, a corresponding multi-beam desired radiation pattern is constructed; Based on the array element parameters and desired beam parameters, each beam is sorted by center angle and the number of snapshots is allocated according to power weight. A phase template corresponding to the beam center is constructed, and the phase offset of each snapshot in the group is generated by indexing the group number using a fixed codebook. After superimposing the phase template and the phase offset, the teacher phase matrix is ​​obtained after physical constraint optimization with the desired multi-beam radiation pattern as the optimization target. The multi-beam desired radiation pattern is input into the phase decoding network. In the phase decoding network, the encoder first extracts global features, and then the group query decoder predicts the group phase template, the snapshot offset within the group, the group size and the group activation state of each beam group, and then obtains the element phase matrix. The teacher phase distillation loss is calculated based on the array element phase matrix and the teacher phase matrix. The self-supervised physical loss, including the pattern loss and the related sidelobe loss, is calculated after substituting the array element phase matrix into the array physical forward model. The phase decoding network is jointly trained using the self-supervised physical loss and the teacher phase distillation loss to obtain the trained phase decoding network. The new multi-beam desired radiation pattern to be processed is input into the trained phase decoding network to obtain the transmit phase matrix of each snapshot of each element on the digital array radar, and a constant envelope transmit waveform is generated based on the transmit phase matrix.

[0105] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0106] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0107] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for rapid generation of multi-beam waveforms in a digital array radar, characterized in that, The method includes: Obtain the array element parameters of the digital array radar, as well as the desired beam parameters during training; Based on the array element parameters and the desired beam parameters, a corresponding multi-beam desired radiation pattern is constructed; Based on the array element parameters and desired beam parameters, each beam is sorted by center angle and the number of snapshots is allocated according to power weight. A phase template corresponding to the beam center is constructed, and the phase offset of each snapshot in the group is generated by indexing the group number using a fixed codebook. After superimposing the phase template and the phase offset, the teacher phase matrix is ​​obtained after physical constraint optimization with the desired multi-beam radiation pattern as the optimization target. The multi-beam desired radiation pattern is input into the phase decoding network. In the phase decoding network, the encoder first extracts global features, and then the group query decoder predicts the group phase template, the snapshot offset within the group, the group size and the group activation state of each beam group, and then obtains the element phase matrix. The teacher phase distillation loss is calculated based on the array element phase matrix and the teacher phase matrix. The self-supervised physical loss, including the pattern loss and the related sidelobe loss, is calculated after substituting the array element phase matrix into the array physical forward model. The phase decoding network is then jointly trained using the self-supervised physical loss and the teacher phase distillation loss to obtain the trained phase decoding network. The new multi-beam desired radiation pattern to be processed is input into the trained phase decoding network to obtain the transmit phase matrix of each snapshot of each element on the digital array radar, and a constant envelope transmit waveform is generated based on the transmit phase matrix.

2. The method for rapid generation of multi-beam waveforms for digital array radar according to claim 1, characterized in that, Based on the array element parameters and the desired beam parameters, the corresponding multi-beam desired radiation pattern is constructed as follows: For each desired beam, the physically realizable main lobe region of the desired beam is determined based on the number of array elements and the spacing between array elements in the array element parameters. Only the power value within the main lobe region is retained, and the desired power outside the main lobe region is set to zero. The main lobe power of each beam is synthesized and normalized according to the power weight in the desired beam parameters to obtain the desired multi-beam radiation pattern.

3. The method for rapid generation of multi-beam waveforms in digital array radar according to claim 2, characterized in that, The physical realization is that the main lobe region is the first zero-point main lobe region corresponding to the beam center, and the boundary of the first zero-point main lobe region is determined by the product of the number of array elements and the array element spacing. If the main lobe regions of multiple beams overlap, the maximum value of the normalized weighted power of each beam is taken as the expected power at the corresponding angle.

4. The method for rapid generation of multi-beam waveforms for digital array radar according to claim 1, characterized in that, The input to the phase decoding network also includes the decibel value sequence and angle Fourier features of the multi-beam desired pattern; The encoder is a Transformer encoder. The phase decoding network extracts the beam center, peak value, width, and activation mask from the multi-beam desired pattern as a pattern descriptor, and fuses the pattern descriptor with the group query features. The group query decoder obtains the context information of each beam group from the global features output by the encoder through a cross-attention mechanism, and predicts the parameters of each beam group based on the context information and the radiation pattern descriptor.

5. The method for rapid generation of multi-beam waveforms for digital array radar according to claim 1, characterized in that, The teacher phase distillation loss includes group phase template loss, group snapshot offset loss, group size loss, group activation loss, and absolute phase loss, which are used to guide the output of the phase decoding network to approximate the generation rule of the teacher phase matrix.

6. The method for rapid generation of multi-beam waveforms for digital array radar according to claim 1, characterized in that, The pattern loss includes pattern main lobe shape loss and main lobe spurious peak suppression loss. The pattern main lobe shape loss is used to constrain the actual transmit pattern corresponding to the network output phase to be consistent with the multi-beam desired pattern in the main lobe region. The main lobe spurious peak suppression loss is used to suppress undesired radiation outside the main lobe region. The correlation sidelobe loss includes autocorrelation peak sidelobe loss and cross-correlation peak sidelobe loss. The autocorrelation peak sidelobe loss is used to constrain the autocorrelation sidelobe level of the same array element waveform under different delays, and the cross-correlation peak sidelobe loss is used to constrain the cross-correlation sidelobe level between different array element waveforms.

7. The method for rapid generation of multi-beam waveforms for digital array radar according to any one of claims 1-6, characterized in that, The joint training uses both fixed-beam-number anchor point samples and random multi-beam samples for training. The fixed beam number anchor point sample is: according to the number of beams in the desired beam parameters, take each fixed integer value from 1 to the preset maximum number of beam groups, and set the center angle and power weight of each beam for each beam number to generate the corresponding multi-beam desired radiation pattern and teacher phase matrix as a training sample pair; The random multi-beam sample is generated by randomly selecting values ​​within their respective ranges according to the number of beams, the center angle of each beam, and the power weight of each beam in the desired beam parameters, and using them as a training sample pair. The joint training includes a fixed-beam-number anchor training phase and a random multi-beam generalization training phase. In the fixed-beam-number anchor training phase, corresponding multi-beam expected radiation patterns and teacher phase matrices are generated for fixed-beam-number anchor samples, and the phase decoding network is trained using teacher phase distillation loss and self-supervised physical loss. In the random multi-beam generalization training phase, the number of beams, beam center angles, and power weights are randomly generated to construct random multi-beam expected radiation patterns, and the phase decoding network is fine-tuned using the self-supervised physical loss. At the same time, fixed-beam-number anchor samples are added to suppress the network's forgetting of existing beam configurations.

8. A device for rapid generation of multi-beam waveforms for digital array radar, characterized in that, The device includes: The training parameter acquisition module is used to acquire the array element parameters of the digital array radar, as well as the desired beam parameters during training. A multi-beam desired pattern generation module is used to construct a corresponding multi-beam desired pattern based on the array element parameters and desired beam parameters. The teacher phase matrix generation module is used to sort each beam by center angle and allocate the number of snapshots according to power weight based on the array element parameters and desired beam parameters, construct a phase template with spatial guidance corresponding to the beam center, and generate the phase offset of each snapshot in the group by group number index using a fixed codebook. After superimposing the phase template and the phase offset, the teacher phase matrix is ​​obtained after physical constraint optimization with the desired multi-beam radiation pattern as the optimization target. The network output module is used to input the multi-beam desired radiation pattern into the phase decoding network. In the phase decoding network, the encoder first extracts global features, and then the group query decoder predicts the group phase template, the snapshot offset within the group, the group size and the group activation state of each beam group, and then obtains the array element phase matrix. The network training module is used to calculate the teacher phase distillation loss based on the array element phase matrix and the teacher phase matrix, and to calculate the self-supervised physical loss including the pattern loss and the related sidelobe loss after substituting the array element phase matrix into the array physical forward model. The self-supervised physical loss and the teacher phase distillation loss are used to jointly train the phase decoding network to obtain the trained phase decoding network. The multi-beam waveform rapid generation module is used to input the new multi-beam desired radiation pattern to be processed into the trained phase decoding network to obtain the transmit phase matrix of each snapshot of each element on the digital array radar, and generate a constant envelope transmit waveform based on the transmit phase matrix.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images