Multi-target doa estimation method based on dbf+CNN

Abstract

The application discloses a multi-target direction of arrival estimation method based on DBF+CNN, and relates to the field of radio direction finding; the method comprises the steps of constructing a DBF+CNN multi-target direction of arrival estimation model, orthogonal double-channel sampling, digital beam forming and realizing spatial filtering, convolution and pooling, and multi-target direction of arrival estimation, etc.; the method utilizes the digital beam forming technology to realize spatial filtering and guiding, and then utilizes a convolution neural network model to realize high-precision estimation of the direction of arrival of multiple targets at the same time on the basis of learning and training; the precision of the direction of arrival of multiple targets at the same time is high, and the direction finding precision is better than 0.15 degrees.

Description

Technical Field

[0001] This invention relates to the field of radio direction finding, specifically a multi-target angle of arrival estimation method based on DBF+CNN. Background Technology

[0002] Angle of arrival (AHA) estimation of radiation source signals is a process used by electronic reconnaissance equipment to estimate the direction of arrival of the radiation source signal by processing the received target signal. Methods for AHA estimation of radiation source signals mainly include amplitude-based direction finding, phase-based direction finding, and spatial spectrum estimation direction finding. These methods are all traditional, non-intelligent AHA estimation methods.

[0003] Chinese patent document CN111610488A proposes an array signal angle of arrival estimation method based on convolutional neural network (CNN). This method, for any given multi-element array, extracts feature information such as the phase difference between array elements from the sampled data of each array element, and achieves fast and high-precision direction finding of radiation source targets based on deep learning.

[0004] However, the aforementioned intelligent angle-of-arrival (AOA) estimation method can only perform direction finding for a single radiation source target. In environments with multiple targets simultaneously, direction finding becomes difficult. This is because: First, this method is an AOA estimation method based on classification and recognition. For a single target, assuming a required direction finding range of 360° and an angular resolution of 0.1°, the number of angle classifications is [value missing]. Under the same conditions, if direction finding is performed for multiple targets, and the number of targets is unknown (possibly 2, 3, 4, ...), the number of angle classifications becomes [value missing]. Considering frequency variations, the required number of training samples is enormous, making the construction of a sample database almost impossible. Second, this method uses feature parameters such as the phase difference between array elements as input parameters for the CNN. In the case of multiple targets simultaneously, multiple signals overlap, and feature parameters such as the phase difference between array elements change, thus leading to direction finding errors.

[0005] To meet the needs of practical applications and solve the problem of intelligent multi-target angle of arrival estimation, we propose a multi-target angle of arrival estimation method based on DBF+CNN. This method uses digital beamforming (DBF) technology to achieve spatial filtering and guidance, and then uses a convolutional neural network model to achieve high-precision estimation of the angle of arrival of multiple targets simultaneously based on learning and training. Summary of the Invention

[0006] The purpose of this invention is to provide a multi-target angle of arrival estimation method based on DBF+CNN. This method addresses the problem that CNN angle of arrival estimation methods can only achieve single-target direction finding. It utilizes digital beamforming technology to achieve spatial filtering and guidance, and then uses a convolutional neural network model to achieve high-precision estimation of the angle of arrival of multiple targets simultaneously based on learning and training.

[0007] This invention can be achieved through the following technical solution: a multi-target angle of arrival estimation method based on DBF+CNN, which includes the following steps:

[0008] Step 1: Construct a DBF+CNN multi-target angle of arrival estimation model;

[0009] Step 2, Orthogonal Dual-Channel Sampling: Through orthogonal transformation and digital sampling processing, I and Q dual-channel sampling data are obtained;

[0010] Step 3: Digital beamforming and implementation of spatial filtering: Simultaneous multi-beam generation based on DBF to achieve spatial filtering;

[0011] Step 4: Convolution and Pooling: Convolutional layers abstract higher-level features, while pooling layers compress data through downsampling operations, dividing the input feature maps into multiple non-overlapping regions.

[0012] Step 5: Multi-target angle of arrival estimation: Utilizing the DBF+CNN multi-target angle of arrival estimation model, high-precision estimation of multi-target angles of arrival is achieved based on learning and training.

[0013] A further technical improvement of the present invention is that the DBF+CNN multi-target angle of arrival estimation model includes 2L+M input layers, 3 convolutional layers, 3 pooling layers, 1 fully connected layer and 1 Softmax classification layer, where L represents the number of antenna elements and M represents the number of multibeams.

[0014] A further technical improvement of this invention is that: digital beamforming includes simultaneous single-beamforming and simultaneous multi-beamforming, and the output of the digital beamformer is a weighted sum of the observation signals of each array element:

[0015] For simultaneous single-beamforming, let the weighting coefficient of the k-th array element be w. k ,make

[0016] w = [w1w2…w] L ] T If we denote the weighting vector, then the output of the digital beamformer is:

[0017]

[0018] Where A(t) m ) represents t mThe signal amplitude at time λ, where L represents the total number of array elements and λ is the signal wavelength;

[0019] For simultaneous multi-beam formation, multiple weighted sums are applied to the array elements simultaneously, and then the sums are applied separately to form multiple different receiving beams.

[0020] A further technical improvement of the present invention is that, in the multi-target environment in the fifth step, based on the spatial filtering of digital beamforming, the fully connected layer couples the distributed generation features to different spatial angles, and on the basis of learning and training, achieves high-precision estimation of the angle of arrival of multiple targets based on DBF+CNN.

[0021] Compared with the prior art, the present invention has the following beneficial effects:

[0022] 1. The method in this invention adds a digital beamforming network to complete spatial filtering and guidance, achieving intelligent multi-target angle of arrival estimation.

[0023] 2. At the same time, it has high multi-target angle of arrival accuracy and direction finding accuracy better than 0.15 degrees. Attached Figure Description

[0024] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0025] Figure 1 This is a flowchart of the DBF+CNN-based multi-target angle of arrival estimation method of the present invention;

[0026] Figure 2 This is a diagram of the multi-target angle of arrival estimation model based on DBF+CNN of the present invention;

[0027] Figure 3 This is a schematic diagram of the multi-beam digital beamforming principle of the present invention. Detailed Implementation

[0028] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided.

[0029] Please see Figure 1 As shown, the multi-target angle of arrival estimation method based on DBF+CNN includes the following steps:

[0030] Step 1: Construct a DBF+CNN multi-target angle of arrival estimation model. The input data of this model consists of orthogonal dual-channel sampled data and multi-beam formed by digital beamforming.

[0031] Step 2: Orthogonal dual-channel sampling. Through orthogonal transformation and digital sampling processing, I and Q dual-channel sampling data are obtained.

[0032] Step 3: Digital beamforming and spatial filtering. Based on orthogonal dual-channel sampled data, DBF technology is used to generate multiple beams simultaneously to achieve spatial filtering.

[0033] Step 4: Convolution and Pooling. The convolutional layer mainly analyzes each small piece of data in greater depth to abstract higher-level features. The pooling layer compresses the data through downsampling operations to reduce the feature dimension after convolution and divides the input feature map into multiple non-overlapping regions.

[0034] Step 5: Multi-target angle of arrival estimation. Based on spatial filtering using digital beamforming, the fully connected layer couples the distributed generated features to different spatial angles. Based on learning and training, high-precision estimation of multi-target angle of arrival based on DBF+CNN is achieved.

[0035] In step one, the DBF+CNN multi-target angle of arrival estimation model is as follows: Figure 2 As shown, in the model construction, the antenna can be set as an array composed of L array elements, and the number of multiple wave speeds can be set as M, so there are L I-channel sampling data and L Q-channel sampling data respectively; the DBF+CNN multi-target angle of arrival estimation model includes 2L+M input layers, 3 convolutional layers, 3 pooling layers, 1 fully connected layer and a Softmax classification layer. The convolutional layers use the Sigmoid activation function, and each convolutional layer is followed by an average pooling layer. The classification layer uses the cross-entropy loss function.

[0036] In step two, assuming that when a plane wave is incident on the antenna array from direction θ, the analog signals received by each antenna element are as follows:

[0037]

[0038] A(t)ω

[0039] Where is the signal amplitude, and is the signal angular frequency;

[0040] For a uniform linear array with element spacing d, the phase Where λ is the signal wavelength;

[0041] The radiation source signals received by each array element undergo orthogonal transformation and digital sampling processing. The orthogonal transformation yields the I and Q channels of the received signal, and the digital sampling processing yields the I and Q channel sampled data of the received signal. At t m The dual-channel sampling data at time t is:

[0042]

[0043] The 2L sampling data obtained by sampling L array elements are used for digital beamforming. At the same time, the 2L sampling data are also the input data of the CNN.

[0044] In step 3, the output of the digital beamformer is the weighted sum of the observation signals of each array element;

[0045] At the same time, for single-beamforming, let the weighting coefficient of the k-th array element be w k , and let w = [w1 w2 … w L T denote the weighting vector, then the output of the digital beamformer is:

[0046]

[0047] For multi-beamforming, if multiple sets of weighting are performed simultaneously and then added separately, multiple different receiving beams can be formed. The principle of digital beamforming for multi-beams is as Figure 3 shown.

[0048] Suppose M beams (M < L - 1) are generated simultaneously. Then, spatial filtering can be achieved using the M beams. The 2L sampling data and the M beams formed by digital beamforming constitute the input data of the CNN.

[0049] In step 4, the convolutional layer mainly analyzes each small piece of data more deeply to abstract higher-level features. The convolutional layer adopts a local connection method and performs convolutional operations with different convolutional kernels on the input data to achieve non-linear feature mapping of the data, thereby realizing feature extraction;

[0050] Given an input signal matrix x ∈ m×n and a filter w ∈ i×j, the filter sequentially performs local convolutional operations on the input features of the previous layer. Then, the output of the convolution is:

[0051]

[0052] The pooling layer realizes data compression through downsampling operations, which is used to reduce the dimensionality of the features after convolution and reduce the number of neurons required by the network. The pooling process is based on the translational invariance property, divides the input feature map into multiple non-overlapping regions, and through the pooling process, automatically extracts the target angle feature vector.

[0053] In step 5, for a multi-target environment, based on the spatial filtering of digital beamforming, the fully connected layer couples the generated features distributed to different spatial angles, and on the basis of learning and training, realizes high-precision estimation of the multi-target direction of arrival based on DBF+CNN.

[0054] ​To verify the performance of this method, we selected a 17-element uniform linear array with a spacing of half a wavelength and conducted a statistical experiment on multi-target angle of arrival estimation using the method of this invention for multiple radiation source targets with different numbers of targets. The statistical results of the experiment are shown in Table 1.

[0055] Table 1. Angle of arrival estimation results

[0056]

[0057] Experimental results show that the method of the present invention can estimate the angle of arrival (AHA) of multiple targets simultaneously, and the AHA estimation accuracy is high, with the AHA estimation accuracy of multiple targets being better than 0.15 degrees.

[0058] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A multi-target angle of arrival estimation method based on DBF+CNN, characterized in that: The method includes the following steps: Acquire 2L channels of I and Q dual-channel sampling data from L antenna array elements; Based on the 2L I and Q dual-channel sampling data, M simultaneous multi-beams are generated through digital beamforming (DBF) to achieve spatial filtering, where M < L-1; A DBF+CNN multi-target angle of arrival estimation model is used for multi-target angle of arrival estimation. The 2L-channel I and Q dual-channel sampling data and the M-channel simultaneous multi-beam are used as inputs to the DBF+CNN multi-target angle of arrival estimation model. Distributed features are extracted through convolution and pooling operations. Based on the learning and training, the distributed features are coupled to different spatial angles through a fully connected layer to achieve high-precision estimation of multi-target angle of arrival.

2. The multi-target angle of arrival estimation method based on DBF+CNN according to claim 1, characterized in that, The DBF+CNN multi-target angle of arrival estimation model includes 2L+M input layers, 3 convolutional layers, 3 pooling layers, 1 fully connected layer, and 1 Softmax classification layer, where L represents the number of antenna elements and M represents the number of multibeams.

3. The multi-target angle of arrival estimation method based on DBF+CNN according to claim 1, characterized in that, The digital beamforming includes simultaneous single-beamforming and simultaneous multi-beamforming, and the output of the digital beamformer is a weighted sum of the observation signals of each array element: For simultaneous single-beamforming, let the weighting coefficient of the k-th array element be... ,make If we denote the weighting vector, then the output of the digital beamformer is: ; in, express The signal amplitude at time t, where L represents the total number of array elements, λ is the signal wavelength, and d is the element spacing. θ The angle of arrival of the target signal to be estimated; For simultaneous multi-beam formation, multiple weighted sums are applied to the array elements simultaneously, and then the sums are applied separately to form multiple different receiving beams.

Images