Radar signal beamforming method, device and electronic equipment

Abstract

The application provides a beamforming method and device of a radar signal, electronic equipment, a computer readable storage medium and a computer program product; the method comprises: receiving radar signals by a plurality of array elements in an array radar respectively, and calculating a noise covariance matrix; performing direction vector prediction processing on the noise covariance matrix and a signal incidence angle range by a first neural network in a beamforming model to obtain first direction vectors of the plurality of array elements; performing full connection processing on the first direction vectors of the plurality of array elements by a second neural network in the beamforming model to obtain second direction vectors of the plurality of array elements; determining the second direction vectors of the plurality of array elements as optimal direction vectors of the plurality of array elements; and performing weighted summation processing on the radar signals received by the plurality of array elements respectively according to the optimal direction vectors of the plurality of array elements to obtain a beamformed radar signal. Through the application, the calculation efficiency and the accuracy of beamforming can be improved.

Description

Technical Field

[0001] The present application relates to radar technology, and in particular to a radar signal beamforming method, device, electronic device, computer-readable storage medium, and computer program product. Background Art

[0002] Radar, a technology that uses radio waves to detect targets and determine their spatial position, is widely used in fields such as transportation, meteorology, and robotics. However, radar signals often contain noise. For example, in vehicles that use high-precision maps and on-board radar for intelligent driving, the increased number of radars on a vehicle often causes interference between radar signals, as does the proximity of similar intelligent driving vehicles.

[0003] Related technologies typically employ the Minimum Variance Distortionless Response (MVDR) algorithm to denoise and mitigate interference in radar signals. However, the MVDR algorithm requires independent execution each time, resulting in lengthy signal processing times and low computational efficiency. Summary of the Invention

[0004] The present application provides a radar signal beamforming method, apparatus, electronic device, computer-readable storage medium, and computer program product, which can improve computing efficiency and beamforming accuracy.

[0005] The technical solution of this application is achieved as follows:

[0006] The present application provides a radar signal beamforming method, comprising:

[0007] Receiving radar signals respectively through a plurality of array elements in the array radar, and calculating a noise covariance matrix according to the radar signals respectively received by the plurality of array elements;

[0008] performing direction vector prediction processing according to the noise covariance matrix and the signal incidence angle range by a first neural network in the beamforming model to obtain first direction vectors of the plurality of array elements; wherein the first neural network is constructed by solving and calculating an objective function of a minimum variance distortionless response (MVDR) by an alternating direction multiplier method (ADMM);

[0009] Performing full-connection processing on the first direction vectors of the plurality of array elements through a second neural network in the beamforming model to obtain second direction vectors of the plurality of array elements; wherein the second neural network includes a plurality of fully-connected layers;

[0010] determining the second direction vectors of the plurality of array elements as the optimal direction vectors of the plurality of array elements;

[0011] A weighted summation process is performed on the radar signals respectively received by the multiple array elements according to the optimal direction vectors of the multiple array elements to obtain a beamformed radar signal.

[0012] The present application provides a radar signal beamforming device, comprising:

[0013] A receiving module, configured to receive radar signals respectively through a plurality of array elements in the array radar, and calculate a noise covariance matrix based on the radar signals respectively received by the plurality of array elements;

[0014] a prediction module, configured to perform directional vector prediction processing based on the noise covariance matrix and the signal incidence angle range through a first neural network in a beamforming model to obtain first directional vectors of the plurality of array elements; wherein the first neural network is constructed by solving and constructing an objective function of a minimum variance distortionless response (MVDR) using an alternating direction multiplier method (ADMM);

[0015] a fully connected module, configured to perform fully connected processing on the first direction vectors of the plurality of array elements through a second neural network in the beamforming model to obtain second direction vectors of the plurality of array elements; wherein the second neural network includes a plurality of fully connected layers;

[0016] a determining module, configured to determine the second direction vectors of the plurality of array elements as the optimal direction vectors of the plurality of array elements;

[0017] The weighted summation module is configured to perform weighted summation processing on the radar signals respectively received by the plurality of array elements according to the optimal direction vectors of the plurality of array elements to obtain a beamformed radar signal.

[0018] The present application provides an electronic device, including:

[0019] a memory for storing executable instructions;

[0020] The processor is configured to implement the radar signal beamforming method provided in the present application when executing the executable instructions stored in the memory.

[0021] The present application provides a computer-readable storage medium storing executable instructions for causing a processor to execute instructions to implement the radar signal beamforming method provided in the present application.

[0022] The present application provides a computer program product, which includes executable instructions for causing a processor to execute the instructions to implement the radar signal beamforming method provided in the present application.

[0023] This application has the following beneficial effects:

[0024] The present application receives radar signals respectively by multiple array elements in an array radar, calculates a noise covariance matrix based on the radar signals respectively received by the multiple array elements, performs direction vector prediction processing based on the noise covariance matrix and the signal incident angle range through a first neural network in a beamforming model to obtain first direction vectors of the multiple array elements, performs full connection processing on the first direction vectors of the multiple array elements through a second neural network in the beamforming model to obtain second direction vectors of the multiple array elements, and finally determines the second direction vectors of the multiple array elements as the optimal direction vectors of the multiple array elements. According to the optimal direction vectors of the multiple array elements, weighted summation processing is performed on the radar signals respectively received by the multiple array elements to obtain the radar signals after beamforming, wherein the first neural network is solved and constructed by the alternating direction multiplier method ADMM for the objective function of the minimum variance distortionless response MVDR, and the second neural network includes multiple fully connected layers. Since deep learning (neural network) is used for calculation in this application, computing efficiency can be improved, effective denoising and anti-interference of radar signals can be achieved at the software level, and hardware costs in actual application scenarios can be reduced; at the same time, the number of layers and neurons in the first neural network and the second neural network are large, which can be applicable to various complex array radars. At the same time, the calculated optimal direction vector is also more accurate, which can improve the denoising and anti-interference effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0026] Figure 1 1 is a schematic diagram of the architecture of a radar signal beamforming system provided in an embodiment of the present application;

[0027] Figure 2 This is a schematic diagram of the structure of a terminal device provided in an embodiment of the present application;

[0028] Figure 3A This is a flow chart of a radar signal beamforming method provided by an embodiment of the present application;

[0029] Figure 3B is another flowchart of the radar signal beamforming method provided in an embodiment of the present application;

[0030] Figure 3C is another flowchart of the radar signal beamforming method provided in an embodiment of the present application;

[0031] Figure 4 is a structural schematic diagram of an array radar provided by an embodiment of the present application;

[0032] Figure 5 is a flowchart of a model training phase of a beamforming model provided by an embodiment of the present application;

[0033] Figure 6 is a structural schematic diagram of a beamforming model provided by an embodiment of the present application. DETAILED DESCRIPTION

[0034] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings, and the described embodiments should not be regarded as limiting the present application, and all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0035] In the following description, “some embodiments” are referred to, which describe a subset of all possible embodiments, but it can be understood that “some embodiments” can be the same subset or different subsets of all possible embodiments, and can be combined with each other without conflict. In the following description, the term “a plurality of” refers to at least two.

[0036] In the following description, the term “first\second” is only to distinguish similar objects, and does not represent a specific order of the objects. It can be understood that “first\second” can be interchanged in a specific order or sequence as allowed, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein.

[0037] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs. The terms used herein are only for the purpose of describing the embodiments of the present application and are not intended to limit the present application.

[0038] Before the embodiments of the present application are further described in detail, the terms and phrases involved in the embodiments of the present application are explained, and the terms and phrases involved in the embodiments of the present application are applicable to the following explanations.

[0039] 1) Beamforming: also known as beamforming, spatial filtering, is a signal processing technique for directional transmission and reception of signals using an array. Beamforming technology adjusts the parameters of the array elements in the phase array, so that the signals of certain angles obtain constructive interference, and the signals of other angles obtain destructive interference, thereby achieving the purpose of signal denoising and anti-interference.

[0040] 2) Array radar: Also known as digital array radar, this is a fully digital array antenna radar based on beamforming principles, where both the receive and transmit beams are formed digitally. An array radar may include an antenna array, digital transmit / receive (T / R) components, a clock, a data transmission system, and a digital processor.

[0041] 3) Element: Refers to the individual receivers or sensors that make up the array. In the embodiments of this application, the array elements in the array radar may be antennas, which are used to receive radar signals from the signal source. The number and arrangement of the elements determine the geometry and performance of the array. Each element has its own receiving directivity and response characteristics, namely, the sensitivity and phase characteristics of the received signal. Through the arrangement and processing of the elements, the array can produce different responses to signals from different directions, thereby achieving functions such as signal source positioning, beamforming, and interference suppression.

[0042] 4) Artificial Neural Network (ANN): This model is an abstraction of the human brain's neural network from an information processing perspective. This model is a computational model composed of a large number of interconnected nodes (or neurons). The first and second neural networks involved in the embodiments of this application are both artificial neural network models.

[0043] 5) Minimum Variance Distortionless Response (MVDR): The MVDR algorithm uses the sampling matrix inversion algorithm used in adaptive beamforming. By adding a weight coefficient (i.e., the direction vectors of multiple array elements) between each array element, it maximizes the interference to noise ratio (INR) of the array in the desired direction and minimizes the variance of the output signal, thereby achieving the purpose of suppressing interference signals.

[0044] 6) Alternating Direction Method of Multipliers (ADMM): This algorithm solves decomposable convex optimization problems by equivalently decomposing the objective function of the original problem (such as the objective function of MVDR) into several solvable subproblems. It then solves each subproblem in parallel and finally coordinates the solutions to the subproblems to obtain a global solution to the original problem.

[0045] 7) Signal to Interference plus Noise Ratio (SINR): This refers to the ratio of the signal in a system to the sum of the interference and noise. SINR is a key technical indicator for measuring the quality and reliability of a signal system. A higher SINR is better.

[0046] The present invention provides a radar signal beamforming method, apparatus, electronic device, computer-readable storage medium, and computer program product, which can improve computational efficiency and beamforming accuracy. The following describes exemplary applications of the electronic device provided by the present invention. The electronic device provided by the present invention can be implemented as various types of terminal devices or as a server.

[0047] See also Figure 1 , Figure 1 This is an architectural diagram of a radar signal beamforming system 100 provided in an embodiment of the present application. A terminal device 400 is connected to a server 200 via a network 300, wherein the network 300 may be a wide area network or a local area network, or a combination of the two.

[0048] In some embodiments, taking the electronic device as a terminal device as an example, the radar signal beamforming method provided in the embodiments of the present application can be implemented by the terminal device. Figure 1 In this example, taking terminal device 400 as an on-board device deployed in a vehicle, terminal device 400 can receive radar signals from multiple array elements in the vehicle's array radar and calculate a noise covariance matrix based on the radar signals received by the multiple array elements. A first neural network in a beamforming model performs directional vector prediction processing based on the noise covariance matrix and the signal incidence angle range to obtain first directional vectors for the multiple array elements. The first neural network is constructed by solving and resolving the objective function of minimum variance distortion-free response (MVDR) using the alternating direction multiplier method (ADMM). A second neural network in the beamforming model performs fully connected processing on the first directional vectors for the multiple array elements to obtain second directional vectors for the multiple array elements. The second neural network includes multiple fully connected layers. The second directional vectors for the multiple array elements are determined as the optimal directional vectors for the multiple array elements. Based on the optimal directional vectors for the multiple array elements, a weighted summation is performed on the radar signals received by the multiple array elements to obtain beamformed radar signals. The beamformed radar signals can be used in advanced driver assistance systems (ADAS), such as for target recognition or location positioning.

[0049] In some embodiments, taking the example of an electronic device being a server, the radar signal beamforming method provided in the embodiments of the present application can also be implemented by the server. For example, after receiving radar signals from multiple array elements in a vehicle's array radar, the terminal device 400 can send the parameters of the radar signals received by the multiple array elements to the server 200, so that the server 200 can calculate the noise covariance matrix. Of course, the terminal device 400 can also send the calculated noise covariance matrix to the server 200. The server 200 uses a first neural network to perform directional vector prediction processing based on the noise covariance matrix and the signal incidence angle range to obtain first directional vectors for the multiple array elements. The server 200 then uses a second neural network to fully connect the first directional vectors of the multiple array elements to obtain second directional vectors for the multiple array elements. The server 200 then determines the second directional vectors of the multiple array elements as the optimal directional vectors for the multiple array elements. The server 200 can then send the optimal directional vectors for the multiple array elements to the terminal device 400, so that the terminal device 400 performs weighted summation processing on the radar signals received by the multiple array elements based on the optimal directional vectors for the multiple array elements to obtain the beamformed radar signals.

[0050] In some embodiments, the server 200 may train the first neural network and the second neural network in the beamforming model, and send the trained first neural network and the trained second neural network to the terminal device 400. The terminal device 400 may locally deploy the trained first neural network and the trained second neural network to calculate the optimal direction vectors of multiple array elements.

[0051] In some embodiments, the terminal device 400 may be a signal processor in an array radar, and may also be other devices for implementing beamforming.

[0052] In some embodiments, the terminal device 400 or the server 200 can implement the radar signal beamforming method provided in the embodiments of the present application by running a computer program. The above-mentioned computer program can be an application, module or plug-in in any form.

[0053] In some embodiments, the server 200 can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms. The terminal device 400 can be, but is not limited to, an in-vehicle device, an aircraft, a smartphone, a tablet computer, a laptop computer, a desktop computer, a smart TV, a smart watch, etc. The terminal device and the server can be connected directly or indirectly via wired or wireless communication, which is not limited in the embodiments of the present application.

[0054] Taking the electronic device provided in the embodiment of the present application as an example, it can be understood that in the case where the electronic device is a server, Figure 2 Parts of the structure shown in FIG (such as the user interface, the presentation module, and the input processing module) may be omitted. Figure 2 , Figure 2 is a schematic diagram of the structure of a terminal device 400 provided in an embodiment of the present application. Figure 2 The terminal device 400 shown includes: at least one processor 410, a memory 450, at least one network interface 420, and a user interface 430. The various components in the terminal device 400 are coupled together via a bus system 440. It is understood that the bus system 440 is used to achieve connection and communication between these components. In addition to including a data bus, the bus system 440 also includes a power bus, a control bus, and a status signal bus. However, for the sake of clarity, the bus system 440 is not shown in FIG. Figure 2 Various buses are labeled as bus system 440 .

[0055] The processor 410 can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., where the general-purpose processor can be a microprocessor or any conventional processor, etc.

[0056] The user interface 430 includes one or more output devices 431 that enable presentation of media content, including one or more speakers and / or one or more visual display screens. The user interface 430 also includes one or more input devices 432, including user interface components that facilitate user input, such as a keyboard, mouse, microphone, touch screen display, camera, other input buttons and controls.

[0057] The memory 450 may be removable, non-removable, or a combination thereof. Exemplary hardware devices include solid-state memory, hard drives, optical drives, etc. The memory 450 may optionally include one or more storage devices that are physically remote from the processor 410.

[0058] Memory 450 includes volatile memory or nonvolatile memory, or may include both volatile and nonvolatile memory. Nonvolatile memory may be read-only memory (ROM), and volatile memory may be random access memory (RAM). The memory 450 described in the embodiments of the present application is intended to include any suitable type of memory.

[0059] In some embodiments, the memory 450 can store data to support various operations, examples of which include programs, modules, and data structures, or a subset or superset thereof, as exemplified below.

[0060] Operating system 451, including system programs for processing various basic system services and performing hardware-related tasks, such as the framework layer, core library layer, and driver layer, which are used to implement various basic services and process hardware-based tasks;

[0061] A network communication module 452 for reaching other computing devices via one or more (wired or wireless) network interfaces 420 , exemplary network interfaces 420 including Bluetooth, WiFi, and USB;

[0062] a presentation module 453 for enabling presentation of information via one or more output devices 431 (e.g., a display screen, a speaker, etc.) associated with the user interface 430 (e.g., a user interface for operating peripheral devices and displaying content and information);

[0063] The input processing module 454 is configured to detect one or more user inputs or interactions from one of the one or more input devices 432 and to translate the detected inputs or interactions.

[0064] In some embodiments, the radar signal beamforming device provided in the embodiments of the present application can be implemented in software. Figure 2Radar signal beamforming device 455 stored in memory 450 is shown. This device can be software in the form of a program or plug-in, and includes the following software modules: a receiving module 4551, a prediction module 4552, a fully connected module 4553, a determination module 4554, and a weighted summation module 4555. These modules are logical and can be arbitrarily combined or further separated depending on the functions they implement. The functions of each module are described below.

[0065] The radar signal beamforming method provided in the embodiment of the present application will be described in conjunction with exemplary applications and implementations of the electronic device provided in the embodiment of the present application.

[0066] See also Figure 3A , Figure 3A This is a flow chart of the radar signal beamforming method provided by the embodiment of the present application, which will be combined with Figure 3A The steps shown are explained.

[0067] In step 101, radar signals are received respectively by a plurality of array elements in an array radar, and a noise covariance matrix is ​​calculated according to the radar signals received respectively by the plurality of array elements.

[0068] Here, the array radar includes multiple array elements, each of which receives radar signals transmitted from an external signal source. A noise covariance matrix can be calculated for the radar signals received by each of the multiple array elements. Because the present embodiment utilizes an artificial neural network for calculations, which offers high computational efficiency, compared to solutions provided by related technologies, the present embodiment imposes fewer restrictions on the number of array elements in the array radar and has a wider range of applicability. For example, the present embodiment is applicable to array radars with hundreds of array elements.

[0069] As an example, the present application provides the following embodiments: Figure 4 The schematic diagram of the array radar shown in FIG. 1 shows a radar array comprising N array elements, each of which is used to receive radar signals. The radar signals received by the N array elements are weighted and summed to obtain the beamformed radar signal. Figure 4 The w1 in represents the weight of array element 1 used for weighted summation, and so on.

[0070] In step 102, a first neural network in the beamforming model performs directional vector prediction processing based on the noise covariance matrix and the signal incident angle range to obtain first directional vectors of multiple array elements; wherein the first neural network is constructed by solving the objective function of the minimum variance distortion-free response (MVDR) using the alternating direction multiplier method (ADMM).

[0071] In the embodiment of the present application, a beamforming model is used to determine the optimal direction vectors of multiple array elements, thereby improving the denoising and anti-interference effects. The beamforming model includes a first neural network and a second neural network, which will be described separately below.

[0072] During the construction of the first neural network, the objective function of the minimum variance distortion-free response (MVDR) is solved using the alternating direction multiplier method (ADMM). Specifically, the ADMM decomposes the objective function into several solvable subproblems, which are then solved using the neural network (the first neural network). It is worth noting that the present embodiment does not limit the objective function, but based on the principles of MVDR, the objective function is at least related to the noise covariance matrix and the signal incidence angle range.

[0073] The first neural network then performs directional vector prediction based on the noise covariance matrix and the signal incident angle range. For ease of distinction, the predicted directional vector is named the first directional vector. It is worth noting that the directional vector prediction process is the process of solving the aforementioned subproblems to obtain the optimal solution to the objective function (i.e., the directional vector).

[0074] In some embodiments, assuming that the number of array elements in the array radar is M, the objective function of MVDR can be:

[0075]

[0076] in, Represents the direction vector of multiple array elements, which is the direction vector of the combination of real and imaginary parts, with a dimension of 2M*1. express The transpose of Represents the real-valued noise covariance matrix, with a dimension of 2M*2M; , is Determine the direction vector, represents the signal incident angle, Indicates the signal incident angle range , express The complementary angle of , Represents the median of the signal incident angle range, that is, ; 、 、 as well as These are all pre-set parameters (hyperparameters) used to determine the allowable error range.

[0077] The above objective function is solved by the alternating direction multiplier method ADMM, and the corresponding first neural network can be constructed.

[0078] In step 103, a second neural network in the beamforming model performs full-connection processing on the first direction vectors of the plurality of array elements to obtain second direction vectors of the plurality of array elements; wherein the second neural network includes a plurality of fully-connected layers.

[0079] Based on the obtained first direction vector, the first direction vector is used as the input of the second neural network. The second neural network performs fully connected processing on the first direction vectors of multiple array elements to obtain second direction vectors of multiple array elements. The purpose of fully connected processing is to achieve information integration, thereby further improving the accuracy of the ultimately obtained optimal direction vector. The second neural network includes multiple fully connected layers.

[0080] The embodiment of the present application does not limit the number of fully connected layers and the number of neurons in each fully connected layer. For example, the second neural network may include three fully connected layers, namely Dense Layer 1, Dense Layer 2 and Dense Layer 3, Dense Layer 1 includes 4M neurons, Dense Layer 2 includes 2M neurons, and Dense Layer 3 includes M neurons.

[0081] In some embodiments, an activation function is included between adjacent fully connected layers in the second neural network. The activation function can be used to perform nonlinear mapping on the linear output of the fully connected layer, thereby enhancing the expressive power of the neural network. This embodiment of the present application does not limit the type of activation function; for example, a rectified linear unit (ReLU) function can be used.

[0082] In some embodiments, a normalization layer is included between two adjacent fully connected layers in the second neural network. The normalization layer can improve the convergence of the second neural network and effectively prevent the second neural network from overfitting during training. The present embodiment does not limit the type of normalization layer; for example, a BatchNormalization layer can be used.

[0083] In some embodiments, the activation function and normalization layer described above may be used simultaneously. For example, for two adjacent fully connected layers in the second neural network, such as fully connected layer 1 and fully connected layer 2, an activation function and a normalization layer are included between fully connected layer 1 and fully connected layer 2, and the normalization layer is located after the activation function.

[0084] As an example, the network layer information of the second neural network shown in Table 1 is provided:

[0085] Table 1 Network layer information

[0086]

[0087] In step 104, the second direction vectors of the plurality of array elements are determined as the optimal direction vectors of the plurality of array elements.

[0088] Here, the second direction vector output by the second neural network is determined as the optimal direction vector, so that accurate and effective beamforming can be achieved according to the optimal direction vector.

[0089] The embodiments of the present application can be applied to various application scenarios of array radar. For example, many vehicles are currently equipped with modules related to intelligent driving, and vehicles that rely entirely on intelligent driving are foreseeable. Vehicles that achieve intelligent driving based on high-precision maps and on-board radars will also become mainstream. Therefore, the embodiments of the present application can be applied to vehicles equipped with array radars, so that the vehicles can accurately receive external radar signals and achieve accurate perception of the external environment, thereby planning reasonable driving strategies to achieve intelligent driving.

[0090] In step 105, weighted summation processing is performed on the radar signals respectively received by the plurality of array elements according to the optimal direction vectors of the plurality of array elements to obtain a beamformed radar signal.

[0091] On the basis of determining the optimal direction vectors of multiple array elements, the weights corresponding to the multiple array elements can be determined (such as Figure 4 w1, w2...w shown N ), and perform weighted summation processing on the radar signals received by multiple array elements according to their corresponding weights to obtain the radar signal after beamforming, so as to effectively achieve denoising and anti-interference.

[0092] like Figure 3AAs shown, in an embodiment of the present application, radar signals are received by multiple array elements in an array radar, and a noise covariance matrix is ​​calculated based on the radar signals received by the multiple array elements. A first neural network in a beamforming model performs direction vector prediction processing based on the noise covariance matrix and the signal incidence angle range to obtain first direction vectors of the multiple array elements. A second neural network in the beamforming model performs full connection processing on the first direction vectors of the multiple array elements to obtain second direction vectors of the multiple array elements. Finally, the second direction vectors of the multiple array elements are determined as the optimal direction vectors of the multiple array elements. Based on the optimal direction vectors of the multiple array elements, weighted sum processing is performed on the radar signals received by the multiple array elements to obtain beamformed radar signals. First, since the embodiment of the present application adopts a deep learning (neural network) method for calculation, compared with the MVDR algorithm, it can improve the computing efficiency, realize effective denoising and anti-interference of radar signals at the software level, and reduce the hardware cost in actual application scenarios; second, the number of layers and neurons in the first neural network and the second neural network is large, which can be applied to various complex array radars. At the same time, the calculated optimal direction vector is more accurate, which can improve the accuracy of beamforming; third, the input of the embodiment of the present application only includes the radar signal and the signal incident angle range, which greatly reduces the prior conditions, has a wider scope of application, and greatly enhances the robustness.

[0093] In some embodiments, see Figure 3B , Figure 3B This is a flow chart of a radar signal beamforming method provided by an embodiment of the present application, based on Figure 3A Before step 101, in step 201, sample radar signals may be received respectively by multiple array elements in the array radar, and a sample noise covariance matrix may be calculated based on the sample radar signals received respectively by the multiple array elements.

[0094] Here, the beamforming model may be trained in advance, so that the optimal direction vector determined by the trained beamforming model is more accurate.

[0095] For example, a specific transmitter can transmit a sample radar signal (i.e., a radar signal used as a sample), which can then be received by multiple elements in an array radar (serving as a receiver). In this way, data samples can be constructed based on the sample radar signals received by the multiple elements in the array radar and the range of their incident angles, and the model training phase can be initiated based on these data samples. It is worth noting that the sample radar signal transmitted by the transmitter can include a desired signal at a specific angle and several (e.g., two) interference signals of the same power. This means that the transmitted sample radar signal contains noise and interference, allowing for testing the denoising and anti-interference performance of the beamforming model.

[0096] After the sample radar signals are received by the plurality of array elements respectively, a sample noise covariance matrix is ​​calculated according to the sample radar signals respectively received by the plurality of array elements.

[0097] In step 202, a first neural network in a beamforming model performs direction vector prediction processing according to a sample noise covariance matrix and a sample signal incident angle range to obtain sample first direction vectors of a plurality of array elements.

[0098] Similar to step 102, the first neural network in the beamforming model performs direction vector prediction processing based on the sample noise covariance matrix and the sample signal incident angle range. For ease of distinction, the direction vector obtained by the direction vector prediction processing here is named the sample first direction vector.

[0099] In step 203, a second neural network in the beamforming model performs full connection processing on the sample first direction vectors of the plurality of array elements to obtain the sample second direction vectors of the plurality of array elements.

[0100] Similarly, similar to step 103, sample second direction vectors of multiple array elements can be obtained.

[0101] In step 204, the sample second direction vectors of the plurality of array elements are determined as the sample optimal direction vectors of the plurality of array elements.

[0102] In step 205, a signal to interference plus noise ratio (SINR) is determined based on the sample optimal direction vectors, the sample noise covariance matrix, and the expected signal power of the plurality of array elements.

[0103] Here, the signal-to-interference-plus-noise ratio (SINR) is determined by combining the sample optimal direction vector, the sample noise covariance matrix, and the expected signal of interest (SOI) power, wherein the SOI power may be estimated.

[0104] The calculation formula for SINR is as follows:

[0105]

[0106] in, represents the expected signal power; Represents the optimal direction vector of the sample; represents the sample noise covariance matrix.

[0107] In step 206, the first neural network and the second neural network in the beamforming model are trained according to the signal to interference noise ratio.

[0108] The signal-to-interference-plus-noise ratio (SINR) is used to evaluate the reliability of signal quality. Specifically, a higher SINR indicates better signal quality. In embodiments of the present application, the SINR can be used to evaluate the quality of a beamforming model. Specifically, a higher SINR indicates better denoising and anti-interference performance, and a more accurate output direction vector. Therefore, the beamforming model can be trained based on the SINR, resulting in a trained beamforming model with improved denoising and anti-interference performance.

[0109] After completing the model training, steps 101 to 105 may be executed, wherein when executing step 102, the trained first neural network is used; and when executing step 103, the trained second neural network is used.

[0110] It is worth noting that the array radar involved in steps 201 to 206 and the array radar involved in steps 101 to 105 can be the same array radar or different array radars.

[0111] In some embodiments, the above-mentioned training of the first neural network and the second neural network in the beamforming model based on the signal-to-interference-noise ratio can be achieved in the following manner: substituting the signal-to-interference-noise ratio into the loss function to obtain a loss value; wherein the loss value is negatively correlated with the signal-to-interference-noise ratio; and updating the model parameters of the first neural network in the beamforming model and the model parameters of the second neural network in the beamforming model based on the loss value.

[0112] Here, the signal-to-interference-noise ratio can be substituted into the loss function to obtain the loss value. There is no restriction on the type of loss function, but it must satisfy the constraint that the loss value is negatively correlated with the signal-to-interference-noise ratio. For example, the loss function can be:

[0113]

[0114] Then, the model parameters of the beamforming model (ie, the model parameters of the first neural network and the model parameters of the second neural network) are updated according to the obtained loss value. For example, a back propagation algorithm may be used to implement the update of the model parameters.

[0115] In some embodiments, the above-mentioned updating of the model parameters of the first neural network in the beamforming model and the model parameters of the second neural network in the beamforming model according to the loss value can be achieved in the following way: the loss values ​​corresponding to multiple data samples are averaged to obtain an average loss value, and the model parameters of the first neural network in the beamforming model and the model parameters of the second neural network in the beamforming model are updated according to the average loss value.

[0116] Each time the transmitter transmits a sample radar signal, it generates a data sample, and each data sample has a corresponding loss value. Therefore, the loss values ​​corresponding to multiple data samples can be averaged to obtain an average loss value. This average loss value is then used to update the beamforming model parameters. This reduces the training workload while ensuring effective model training, facilitating rapid iteration.

[0117] It is worth noting that in order to improve the model training effect, different sample radar signals can be transmitted by the transmitter at different times.

[0118] In some embodiments, multiple data samples can be divided into a training set and a test set; a beamforming model is trained based on the training set; model indicators of the trained beamforming model are determined based on the test set; and whether to continue training the beamforming model is determined based on the model indicators.

[0119] As an example, the present application provides the following embodiments: Figure 5 The flow chart of the model training phase shown in the figure will be combined with Figure 5 The steps shown are explained.

[0120] Step 1: Construct a data set and divide it into a training set and a test set. There is no limit on the division method. For example, the training set includes 10 7 data samples, the test set includes 10 6 data samples.

[0121] Step 2: Construct a beamforming model based on the TensorFlow deep learning framework. This step enables the normal use of the beamforming model. Of course, the TensorFlow deep learning framework used here is only an example and does not constitute a limitation on the embodiments of this application.

[0122] Step 3: Input the training set obtained in step 1 into the beamforming model to train the beamforming model.

[0123] Step 4: Determine the model metric of the trained beamforming model based on the test set obtained in Step 1. If the model metric is greater than the metric threshold, training is terminated. If the model metric is less than or equal to the metric threshold, step 3 is repeated. The model metric may be the signal-to-interference-and-noise ratio.

[0124] Through the above method, it is possible to accurately determine whether the beamforming model is fully trained and end the model training at an appropriate time.

[0125] In some embodiments, multiple data samples can be divided into a training set, a test set, and a validation set; multiple beamforming models are trained separately according to the training set, wherein different beamforming models have different model settings; model indicators of the trained multiple beamforming models are determined according to the validation set, and the optimal model settings are determined according to the model indicators of the multiple beamforming models; the optimal beamforming model is trained according to the training set and the validation set, wherein the optimal beamforming model refers to a beamforming model that applies the optimal model settings; the model indicators of the trained optimal beamforming model are determined according to the test set, and it is determined whether to continue training the optimal beamforming model based on the model indicators.

[0126] Here, we can construct a data set and divide it into a training set, a test set, and a validation set. There is no limit on the division method. For example, the training set includes 10 7 data samples, and the validation set includes 10 6 data samples, the test set includes 10 5 data samples.

[0127] Then, multiple beamforming models are trained separately based on the training set, where different beamforming models have different model settings, where the model settings refer to settings that are unrelated to model parameters during the model training phase, and may include, for example, at least one of hyperparameters, number of iterations, learning rate, activation function type, and normalization method. Model indices of the multiple beamforming models after training are determined based on the validation set, and the model setting corresponding to the largest model indices is used as the optimal model setting. Then, the optimal beamforming model is trained based on the training set and the validation set, that is, the training set and the validation set are used together as a new training set, where the optimal beamforming model refers to a beamforming model that applies the optimal model settings. Model indices of the optimal beamforming model after training are determined based on the test set, and whether to continue training the optimal beamforming model is determined based on the model indices.

[0128] Through the above method, the optimal model settings can be determined to further improve the effect of model training.

[0129] like Figure 3B As shown, unlike the traditional supervised learning network, the embodiment of the present application directly calculates the SINR to train the beamforming model, without spending a lot of time to calculate the supervision value (label) through the traditional MVDR algorithm, which can reduce the cost spent in the model training stage and improve the model training efficiency; at the same time, the beamforming model has a large number of model parameters, strong robustness and stability, and the trained beamforming model has strong denoising and anti-interference performance.

[0130] In some embodiments, see Figure 3C , Figure 3Cis a flow chart of a radar signal beamforming method provided in an embodiment of the present application. Figure 3A Step 102 shown can be updated to step 301, in which K rounds of outer loops are executed through the first neural network in the beamforming model, and N rounds of inner loops are executed in each round of outer loops; wherein, the following processing is performed in the kth round of outer loop: N rounds of inner loops are executed on the noise covariance matrix, the signal incident angle range, and the auxiliary parameters of the kth round of outer loop to obtain the direction vector of the k+1th round of outer loop; based on the direction vector of the k+1th round of outer loop and the signal incident angle range, the auxiliary parameters of the k+1th round of outer loop are determined; wherein, the direction vector of the kth round of outer loop is used as the first direction vector of multiple array elements.

[0131] Here, the beamforming model can be a Multi-Loop Full-digital Neural Network (MLFNN) model. For example, solving the MVDR objective function using ADMM yields several subproblems in the form of nested loops. The nested loops consist of K outer loops, each of which contains N inner loops, where K and N are both integers greater than 1.

[0132] On this basis, K rounds of outer loops are executed through the first neural network, and N rounds of inner loops are executed within each outer loop. For ease of understanding, take the kth outer loop as an example, where k is an integer greater than 0 and less than K, and n in the following text is an integer greater than 0 and less than N. The following processing is performed in the kth outer loop: N rounds of inner loops are executed on the noise covariance matrix, the signal incident angle range, and the auxiliary parameters of the kth outer loop to obtain the direction vector of the k+1th outer loop; based on the direction vector of the k+1th outer loop and the signal incident angle range, the auxiliary parameters of the k+1th outer loop are determined. That is, auxiliary parameters are introduced to implement nested loops.

[0133] It is worth noting that the calculated direction vector of the Kth outer cycle is used as the first direction vector of multiple array elements.

[0134] In some embodiments, the following processing is performed in the kth outer loop:

[0135]

[0136]

[0137] in, represents the noise covariance matrix; Represents the auxiliary parameters of the k-th outer loop; Represents the direction vector of the k+1th external loop, i.e. ; , Represents the median value of the signal incident angle range; represents the Lagrange multiplier; represents the penalty coefficient; , Indicates the signal incident angle range, express The complementary angle of Indicates the signal incident angle.

[0138] in, 、 、 as well as are the model parameters of the first neural network.

[0139] In some embodiments, as well as It varies with the number of rounds k of the outer loop and the number of rounds n of the inner loop; as well as As the number of rounds k of the outer loop changes. Thus, the first neural network has 2 K*N +2 K model parameters. Since the number of model parameters is large, the robustness and stability of the first neural network can be improved.

[0140] like Figure 3C As shown, the embodiment of the present application solves the sub-problem through the nested loop in the first neural network, which is equivalent to solving the objective function of MVDR and obtaining the first direction vector as the optimal solution.

[0141] The present application provides an embodiment of Figure 6 The structural diagram of the beamforming model shown in FIG corresponds to the model training phase. In the model training phase, the input of the beamforming model includes the sample noise covariance matrix Additive noise signal , estimated sample signal incident angle range and the estimated desired signal power The output of the beamforming model is the signal-to-interference-plus-noise ratio (SIN), which is calculated based on the direction vector (second direction vector) output by the second neural network. During the model inference phase following model training, the output of the beamforming model is the optimal direction vector, which is the second direction vector output by the second neural network.

[0142] The following continues to describe an exemplary structure of the radar signal beamforming device 455 provided in the embodiment of the present application implemented as a software module. In some embodiments, such as Figure 2 As shown, the software modules stored in the beamforming device 455 of the radar signal in the memory 450 can include: a receiving module 4551, configured to receive radar signals by a plurality of array elements in the array radar respectively, and calculate a noise covariance matrix according to the radar signals received by the plurality of array elements respectively; a prediction module 4552, configured to perform direction vector prediction processing on the noise covariance matrix and a signal incident angle range by a first neural network in a beamforming model, to obtain a first direction vector of the plurality of array elements; wherein the first neural network is constructed by solving a target function of a minimum variance distortion response (MVDR) through an alternating direction multiplier method (ADMM); a full connection module 4553, configured to perform full connection processing on the first direction vector of the plurality of array elements by a second neural network in the beamforming model, to obtain a second direction vector of the plurality of array elements; wherein the second neural network includes a plurality of full connection layers; a determination module 4554, configured to determine the second direction vector of the plurality of array elements as an optimal direction vector of the plurality of array elements; and a weighted sum module 4555, configured to perform weighted sum processing on the radar signals received by the plurality of array elements respectively according to the optimal direction vector of the plurality of array elements, to obtain a beamformed radar signal.

[0143] In some embodiments, the beamforming device 455 of the radar signal further includes a training module, configured to: receive sample radar signals by the plurality of array elements respectively, and calculate a sample noise covariance matrix according to the sample radar signals received by the plurality of array elements respectively; perform direction vector prediction processing on the sample noise covariance matrix and a sample signal incident angle range by the first neural network in the beamforming model, to obtain a sample first direction vector of the plurality of array elements; perform full connection processing on the sample first direction vector of the plurality of array elements by the second neural network in the beamforming model, to obtain a sample second direction vector of the plurality of array elements; determine the sample second direction vector of the plurality of array elements as a sample optimal direction vector of the plurality of array elements; determine a signal-to-interference-plus-noise ratio according to the sample optimal direction vector of the plurality of array elements, the sample noise covariance matrix and an expected signal power; and train the first neural network and the second neural network in the beamforming model according to the signal-to-interference-plus-noise ratio.

[0144] In some embodiments, the training module is further configured to: substitute the signal-to-interference-plus-noise ratio into a loss function to obtain a loss value; wherein the loss value is negatively correlated with the signal-to-interference-plus-noise ratio; and update model parameters of the first neural network in the beamforming model and model parameters of the second neural network in the beamforming model according to the loss value.

[0145] In some embodiments, the prediction module 4552 is further used to execute K rounds of outer loops through the first neural network in the beamforming model, and to execute N rounds of inner loops in each round of outer loops; wherein, the following processing is performed in the kth round of outer loops: N rounds of inner loops are executed on the noise covariance matrix, the signal incident angle range, and the auxiliary parameters of the kth round of outer loops to obtain the direction vector of the k+1th round of outer loops; based on the direction vector of the k+1th round of outer loops and the signal incident angle range, the auxiliary parameters of the k+1th round of outer loops are determined; wherein, the direction vector of the Kth round of outer loops is used as the first direction vector of multiple array elements.

[0146] In some embodiments, the following processing is performed in the kth outer loop:

[0147]

[0148]

[0149] in, represents the noise covariance matrix; Represents the auxiliary parameters of the k-th outer loop; Represents the direction vector of the k+1th external loop, i.e. ; , Represents the median value of the signal incident angle range; represents the Lagrange multiplier; represents the penalty coefficient; , Indicates the signal incident angle range, express The complementary angle of Indicates the signal incident angle;

[0150] in, 、 、 as well as are the model parameters of the first neural network.

[0151] In some embodiments, as well as It varies with the number of rounds k of the outer loop and the number of rounds n of the inner loop; as well as Varies with the number of rounds k in the outer loop.

[0152] In some embodiments, an activation function and a normalization layer are further included between two adjacent fully connected layers in the second neural network.

[0153] Embodiments of the present application provide a computer program product or computer program, which includes executable instructions stored in a computer-readable storage medium. A processor of an electronic device reads the executable instructions from the computer-readable storage medium and executes the executable instructions, causing the electronic device to perform the radar signal beamforming method described in the embodiments of the present application.

[0154] An embodiment of the present application provides a computer-readable storage medium storing executable instructions. The computer-readable storage medium stores executable instructions that, when executed by a processor, cause the processor to execute the radar signal beamforming method provided by the embodiment of the present application.

[0155] In some embodiments, the computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash memory, magnetic surface storage, optical disk, or CD-ROM; or various devices including one or any combination of the above memories.

[0156] In some embodiments, executable instructions may be in the form of a program, software, software module, script, or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a stand-alone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0157] As an example, executable instructions may, but need not, correspond to a file in a file system, may be stored as part of a file that stores other programs or data, such as in one or more scripts in a HyperText Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple coordinating files (e.g., files storing one or more modules, subroutines, or code portions).

[0158] By way of example, executable instructions may be deployed to be executed on one computing device, or on multiple computing devices at one site, or on multiple computing devices distributed across multiple sites and interconnected by a communication network.

[0159] The above are merely examples of the present application and are not intended to limit the scope of protection of the present application. Any modifications, equivalent replacements, and improvements made within the spirit and scope of the present application are included in the scope of protection of the present application.

Claims

1. A radar signal beamforming method, characterized in that: include: Receiving radar signals respectively through a plurality of array elements in the array radar, and calculating a noise covariance matrix according to the radar signals respectively received by the plurality of array elements; performing direction vector prediction processing according to the noise covariance matrix and the signal incidence angle range by a first neural network in the beamforming model to obtain first direction vectors of the plurality of array elements; wherein the first neural network is constructed by solving and calculating an objective function of a minimum variance distortionless response (MVDR) by an alternating direction multiplier method (ADMM); Performing full-connection processing on the first direction vectors of the plurality of array elements through a second neural network in the beamforming model to obtain second direction vectors of the plurality of array elements; wherein the second neural network includes a plurality of fully-connected layers; determining the second direction vectors of the plurality of array elements as the optimal direction vectors of the plurality of array elements; A weighted summation process is performed on the radar signals respectively received by the multiple array elements according to the optimal direction vectors of the multiple array elements to obtain a beamformed radar signal.

2. The method according to claim 1, characterized in that Before respectively receiving radar signals by a plurality of array elements in the array radar, the method further includes: Receiving sample radar signals respectively through the plurality of array elements, and calculating a sample noise covariance matrix according to the sample radar signals respectively received by the plurality of array elements; performing direction vector prediction processing according to the sample noise covariance matrix and the sample signal incident angle range by the first neural network in the beamforming model to obtain sample first direction vectors of the multiple array elements; Performing full-connection processing on the sample first direction vectors of the plurality of array elements through the second neural network in the beamforming model to obtain the sample second direction vectors of the plurality of array elements; Determining the sample second direction vectors of the plurality of array elements as the sample optimal direction vectors of the plurality of array elements; Determining a signal to interference plus noise ratio according to the sample optimal direction vectors of the plurality of array elements, the sample noise covariance matrix, and the expected signal power; The first neural network and the second neural network in the beamforming model are trained according to the signal to interference plus noise ratio.

3. The method according to claim 2, characterized in that The training of the first neural network and the second neural network in the beamforming model according to the signal to interference and noise ratio includes: Substituting the signal to interference noise ratio into a loss function to obtain a loss value; wherein the loss value is negatively correlated with the signal to interference noise ratio; Model parameters of the first neural network in the beamforming model and model parameters of the second neural network in the beamforming model are updated according to the loss value.

4. The method according to any one of claims 1 to 3, characterized in that The performing direction vector prediction processing according to the noise covariance matrix and the signal incident angle range by the first neural network in the beamforming model to obtain the first direction vectors of the multiple array elements includes: executing K outer loops through the first neural network in the beamforming model and executing N inner loops within each outer loop; The following processing is performed in the kth outer loop: Performing N rounds of inner loops on the noise covariance matrix, the signal incident angle range, and the auxiliary parameters of the k-th outer loop to obtain a direction vector for the k+1-th outer loop; Determining auxiliary parameters of the k+1th round of external loop according to the direction vector of the k+1th round of external loop and the signal incidence angle range; The direction vector of the Kth outer cycle is used as the first direction vector of the multiple array elements.

5. The method according to claim 4, characterized in that In the kth round of the outer loop, the following processing is performed: in, represents the noise covariance matrix; Represents the auxiliary parameters of the k-th outer loop; Represents the direction vector of the k+1th external loop, i.e. ; , represents the median value of the signal incident angle range; represents the Lagrange multiplier; represents the penalty coefficient; , represents the signal incident angle range, express The complementary angle of represents the signal incident angle, is Determine the direction vector; in, 、 、 as well as are all model parameters of the first neural network.

6. The method according to claim 5, characterized in that as well as It varies with the number of rounds k of the outer loop and the number of rounds n of the inner loop; as well as Varies with the number of rounds k in the outer loop.

7. The method according to any one of claims 1 to 3, characterized in that An activation function and a normalization layer are also included between two adjacent fully connected layers in the second neural network.

8. A radar signal beamforming device, characterized in that: include: A receiving module, configured to receive radar signals respectively through a plurality of array elements in the array radar, and calculate a noise covariance matrix based on the radar signals respectively received by the plurality of array elements; a prediction module, configured to perform directional vector prediction processing based on the noise covariance matrix and the signal incidence angle range through a first neural network in a beamforming model to obtain first directional vectors of the plurality of array elements; wherein the first neural network is constructed by solving and constructing an objective function of a minimum variance distortionless response (MVDR) using an alternating direction multiplier method (ADMM); a fully connected module, configured to perform fully connected processing on the first direction vectors of the plurality of array elements through a second neural network in the beamforming model to obtain second direction vectors of the plurality of array elements; wherein the second neural network includes a plurality of fully connected layers; a determining module, configured to determine the second direction vectors of the plurality of array elements as the optimal direction vectors of the plurality of array elements; The weighted summation module is configured to perform weighted summation processing on the radar signals respectively received by the plurality of array elements according to the optimal direction vectors of the plurality of array elements to obtain a beamformed radar signal.

9. An electronic device, characterized in that: include: a memory for storing executable instructions; A processor, configured to implement the radar signal beamforming method according to any one of claims 1 to 7 when executing the executable instructions stored in the memory.

10. A computer-readable storage medium, characterized in that Executable instructions are stored, and when executed by a processor, the radar signal beamforming method according to any one of claims 1 to 7 is implemented.

Images