CN-121980378-A - Broadband incoming wave direction estimation method based on convolutional neural network

CN121980378ACN 121980378 ACN121980378 ACN 121980378ACN-121980378-A

Abstract

The invention belongs to the technical field of signal processing, and particularly discloses a broadband incoming wave direction estimation method and system based on a convolutional neural network. The method comprises the steps of constructing a broadband incoming wave direction estimation model based on a convolutional neural network, wherein the broadband incoming wave direction estimation model is used for calculating posterior probability of an angle of a broadband incoming wave direction of an input signal, training the broadband incoming wave direction estimation model by using a phase part of a synthesized signal and a corresponding label as training data, and obtaining the broadband incoming wave direction estimation angle output by the broadband incoming wave direction estimation model by taking the phase part of the incident signal as input of the trained broadband incoming wave direction estimation model. The method solves the technical problems of high calculation cost and poor robustness existing in the existing incoming wave direction estimation.

Inventors

Request for anonymity
Request for anonymity
Request for anonymity
Request for anonymity

Assignees

北京遥感设备研究所

Dates

Publication Date: 20260505
Application Date: 20251223

Claims (10)

1. The broadband incoming wave direction estimation method based on the convolutional neural network is characterized by comprising the following steps of: constructing a broadband incoming wave direction estimation model based on a convolutional neural network, wherein the broadband incoming wave direction estimation model is used for calculating posterior probability of an angle of a broadband incoming wave direction of an input signal; Training the broadband incoming wave direction estimation model by using the phase part of the synthesized signal and the corresponding label as training data; and taking the phase part of the incident signal as the input of the broadband incoming wave direction estimation model after training to obtain the broadband incoming wave direction estimation angle output by the broadband incoming wave direction estimation model.
2. The broadband incoming wave direction estimation method based on the convolutional neural network as set forth in claim 1, wherein constructing the broadband incoming wave direction estimation model based on the convolutional neural network comprises: And constructing a classifier of a broadband incoming wave direction estimation model, wherein the classifier is used for mapping phase characteristic vectors of input signals and various broadband incoming wave direction estimation categories, and each broadband incoming wave direction estimation category corresponds to different angle ranges.
3. The broadband incoming wave direction estimation method based on the convolutional neural network as set forth in claim 2, wherein constructing a classifier of the broadband incoming wave direction estimation model comprises: The method comprises the steps of constructing an input layer, a composite downsampling layer, a composite upsampling layer, a full convolution layer and an output layer which are sequentially connected, wherein the input layer outputs a multidimensional feature map, and the output layer determines a broadband incoming wave direction estimation category corresponding to the maximum value of posterior probability.
4. The broadband incoming wave direction estimation method based on the convolutional neural network as recited in claim 3, wherein constructing a classifier of the broadband incoming wave direction estimation model comprises: the composite downsampling layer comprises a plurality of groups of convolution layers which are sequentially connected, wherein each group of convolution layers comprises a convolution layer, a maximum pooling layer, a batch normalization layer and a ReLU activation function; The composite upsampling layer comprises a plurality of groups of upsampling layers which are sequentially connected, wherein each group of upsampling layers comprises an upsampling layer, a jump connection layer, a batch normalization layer and a ReLU activation function; the full convolution layer comprises at least two convolution layers, wherein the number of channels of the last convolution layer is consistent with the number of broadband incoming wave direction estimation categories.
5. The broadband incoming wave direction estimation method based on the convolutional neural network as set forth in claim 4, wherein constructing a classifier of the broadband incoming wave direction estimation model comprises: the multiple groups of convolution layers of the composite downsampling layer sequentially reduce the size of the input feature map; And a plurality of groups of up-sampling layers of the composite up-sampling layer sequentially increase the size of the input characteristic diagram in a linear difference mode.
6. The broadband incoming wave direction estimation method based on a convolutional neural network according to claim 2, wherein training the broadband incoming wave direction estimation model using the phase portion of the synthesized signal and the corresponding tag as training data comprises: and mapping and training the classifier of the broadband incoming wave direction estimation model by using the phase characteristic vector of the synthesized signal and the corresponding broadband incoming wave direction estimation class label.
7. The broadband incoming wave direction estimation method based on the convolutional neural network according to claim 2, wherein the step of obtaining the broadband incoming wave direction estimation angle output by the broadband incoming wave direction estimation model by taking the phase part of the incident signal as the input of the trained broadband incoming wave direction estimation model comprises the following steps: determining a phase eigenvector formed by a phase portion of the incident signal; Calculating posterior probability of each broadband incoming wave direction category corresponding to the phase characteristic vector; and outputting the angle range of the broadband incoming wave direction category corresponding to the maximum value of the backward delay probability.
8. A broadband incoming wave direction estimation system based on a convolutional neural network, comprising: The construction module is used for constructing a broadband incoming wave direction estimation model based on the convolutional neural network, wherein the broadband incoming wave direction estimation model is used for calculating posterior probability of an angle of a broadband incoming wave direction of an input signal; The training module is used for training the broadband incoming wave direction estimation model by using the phase part of the synthesized signal and the corresponding label as training data; And the application module is used for taking the phase part of the incident signal as the input of the broadband incoming wave direction estimation model after training to obtain the broadband incoming wave direction estimation angle output by the broadband incoming wave direction estimation model.
9. A computer readable storage medium having a computer program stored therein, characterized in that the computer program is arranged to perform the method of any of claims 1 to 7 when run.
10. An electronic device comprising a memory and a processor, the memory having stored therein a computer program, characterized in that the processor is arranged to execute the method according to any of claims 1 to 7 by means of the computer program.

Description

Broadband incoming wave direction estimation method based on convolutional neural network Technical Field The invention belongs to the technical field of signal processing, and particularly relates to a broadband incoming wave direction estimation method based on a convolutional neural network. Background The incoming wave direction estimation, namely DOA estimation, is one of the important directions in the field of array signal processing, and is applied to many fields such as military, communication and the like. DOA estimation is typically accomplished by information of the signal received by the array antenna. Methods for wideband signal DOA estimation have been known for many years, including subspace-based Methods (MUSIC), time difference of arrival-based methods (TPOA), correlation methods (SRPPHAT), model-based methods (maximum likelihood estimation methods), but these methods have problems of high computational cost and poor robustness. Disclosure of Invention The invention aims to provide a broadband incoming wave direction estimation method based on a convolutional neural network, which aims to solve the technical problems of high calculation cost and poor robustness in the existing incoming wave direction estimation. In order to achieve the above purpose, the invention adopts the following technical scheme: A broadband incoming wave direction estimation method based on a convolutional neural network comprises the steps of constructing a broadband incoming wave direction estimation model based on the convolutional neural network, wherein the broadband incoming wave direction estimation model is used for calculating posterior probability of an angle of a broadband incoming wave direction of an input signal, training the broadband incoming wave direction estimation model by using a phase part of a synthesized signal and a corresponding label as training data, and obtaining a broadband incoming wave direction estimation angle output by the broadband incoming wave direction estimation model by using the phase part of an incident signal as input of the broadband incoming wave direction estimation model after training. Preferably, constructing a broadband incoming wave direction estimation model based on a convolutional neural network comprises constructing a classifier of the broadband incoming wave direction estimation model, wherein the classifier is used for mapping phase feature vectors of input signals and various broadband incoming wave direction estimation categories, and each broadband incoming wave direction estimation category corresponds to a different angle range. Preferably, constructing a classifier of a broadband incoming wave direction estimation model comprises constructing an input layer, a composite downsampling layer, a composite upsampling layer, a full convolution layer and an output layer which are sequentially connected, wherein the input layer outputs a multi-dimensional feature map, and the output layer determines a broadband incoming wave direction estimation class corresponding to a maximum posterior probability. The classifier for constructing the broadband incoming wave direction estimation model comprises a plurality of groups of convolution layers which are sequentially connected, wherein each group of convolution layers comprises a convolution layer, a maximum pooling layer, a batch normalization layer and a ReLU activation function, each group of up-sampling layers comprises a plurality of groups of up-sampling layers which are sequentially connected, each group of up-sampling layers comprises an up-sampling layer, a jump connection layer, a batch normalization layer and a ReLU activation function, the total convolution layer comprises at least two convolution layers, and the number of channels of the last convolution layer is consistent with that of the broadband incoming wave direction estimation classes. Preferably, the classifier for constructing the broadband incoming wave direction estimation model comprises a plurality of groups of convolution layers of the composite downsampling layer, wherein the plurality of groups of convolution layers of the composite downsampling layer sequentially reduce the size of an input feature map, and the plurality of groups of upsampling layers of the composite upsampling layer sequentially increase the size of the input feature map in a linear difference mode. Preferably, training the broadband incoming wave direction estimation model by using the phase part of the synthesized signal and the corresponding label as training data comprises mapping training a classifier of the broadband incoming wave direction estimation model by using a phase feature vector of the synthesized signal and the corresponding broadband incoming wave direction estimation class label. Preferably, the phase part of the incident signal is used as the input of the broadband incoming wave direction estimation model after training to obtain the broadband inc