An echo-sounder echo simulation device and method based on array phase weight estimation
Patent Information
- Application Number
- CN202311358783.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-10-19
- Publication Date
- 2026-08-18
- Estimated Expiration
- 2043-10-19
AI Technical Summary
[0003]但是目前声呐回波模拟器的研究主要集中在水声环境的仿真和回波信号的生成，未呈现系统特性；并且部分声呐回波模拟器通过旋转矩阵等技术估计目标位置，该类方法实时性低，误差较大
[0036] 1. The sonar echo simulation device provided by the present invention can autonomously complete high-resolution real-time angle of arrival estimation and generate directional simulated echoes based on array structure, which solves the problem that current echo devices cannot obtain real-time angles and improves the realism of echo simulation.
Smart Images
Figure CN117420535B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of sonar equipment technology, and particularly relates to a sonar echo simulation device and method based on array phase weight estimation. Background Technology
[0002] Sonar identifies and locates underwater targets by receiving reflected sound waves. The complex aquatic environment significantly impacts sonar performance testing, often requiring multiple trials in real-world scenarios. However, the complex marine environment and demanding field testing conditions make it difficult to place targets in suitable locations and under appropriate conditions. Therefore, early-stage sonar development requires support in a laboratory environment. A sonar echo simulator, capable of simulating the return signal from a sonar target, can address these challenges. The sonar echo simulator can simulate the return signal from a sonar target in a water tank testing environment, aiding in sonar development and accelerating the process. When the sonar signal reaches the sonar echo simulator, the direction of arrival is first estimated based on the received signal to obtain the sonar's azimuth. Furthermore, a simulated echo is generated using preset geometric features such as object dimensions and is transmitted towards the sonar.
[0003] However, current research on sonar echo simulators mainly focuses on the simulation of underwater acoustic environments and the generation of echo signals, without revealing system characteristics. Furthermore, some sonar echo simulators estimate target positions using techniques such as rotation matrices, which suffer from low real-time performance and large errors. Therefore, we propose a sonar echo simulation device and method based on array phase weight estimation. Summary of the Invention
[0004] The purpose of this invention is to provide a sonar echo simulation device and method based on array phase weight estimation, which aims to solve the problems mentioned in the background art.
[0005] To achieve the above objectives, the present invention provides the following technical solution:
[0006] A sonar echo simulation device based on array phase weight estimation includes:
[0007] The simulated target information preset module is used to preset target parameters and obtain the transfer function;
[0008] The signal acquisition module is used to receive array signals;
[0009] The array signal filtering module is used to reduce noise in the array signal;
[0010] The sonar signal azimuth estimation module is used to estimate the azimuth of the sonar signal.
[0011] The echo generation module is used to perform reverse processing based on the estimation results to generate the echo of the simulated target;
[0012] The signal transmission module is used to send the echo of the simulated target in the direction of the sonar.
[0013] Furthermore, the specific operation of the sonar signal azimuth estimation module in estimating the sonar signal azimuth is as follows:
[0014] The received signals of each array element are obtained, and the arrival angle direction is estimated based on the beamforming principle and convolutional neural network model. The estimation result is the phase weight of each array element.
[0015] A sonar echo simulation method based on array phase weight estimation includes the following steps:
[0016] Step S1: Preset target parameters and obtain the transfer function: Treat the scattered echo from the target as a superposition of several sub-echoes, and consider the sub-echoes as emitted from bright spots; express the transfer function as:
[0017] H(r,ω)=A(r,ω)e jωτ e jφ ;
[0018] Where r is the distance between the target and the echo receiver, ω is the angular frequency, j is the imaginary unit, τ is the time delay, φ is the phase jump factor, and A(r,ω) is the amplitude factor;
[0019] Step S2: Receive the array signal, reduce noise in the array signal, and estimate the sonar signal azimuth; the array has m elements, the element spacing is d, and the sound wave is incident from angle θ; for a uniform linear array, the time delay difference between adjacent elements is:
[0020]
[0021] Where c is the underwater acoustic speed; the frequency domain phase shift model of the signal received by the array element is as follows:
[0022]
[0023] Further transformed into:
[0024]
[0025] Assume the sound source signal vector is s n (t), where the number of narrowband sound sources is N, n m (k) is the noise vector received by the array element, then the signal collected by the linear array is:
[0026]
[0027] The direction of arrival estimation of the array received signal is transformed into phase weight estimation between each adjacent array element;
[0028] Step S3: Construct an end-to-end neural network model, use CNN to extract features of amplitude and phase and learn data, and feed the real and imaginary parts of the input signal into the end-to-end neural network model for learning respectively;
[0029] Step S4: After obtaining the phase weight results of the array elements through the end-to-end neural network model, the simulated echo is calculated based on the azimuth information input by the transfer function. The obtained phase weight information is then sent back into the array transmitting circuit to realize the end-to-end reception and simulated transmission of sonar echoes.
[0030] Furthermore, in step S3, the specific operation of feeding the real and imaginary parts of the input signal into the end-to-end neural network model for learning is as follows:
[0031] When the number of array elements M = 16 and the number of snapshots is n, the input signal is a real matrix and an imaginary matrix with a size of M*n and a channel number of 1. The input signal is sent to the convolutional layer for feature learning. After the real matrix and the imaginary matrix have completed the convolution operation, dimensionality reduction and feature concatenation are performed respectively, and then sent to two fully connected layers to form a one-dimensional feature vector.
[0032] Furthermore, the convolutional layer includes two branches, each branch including four layers. The first layer includes 32 convolutional kernels, each with a kernel size of 3, and uses max pooling to reduce the data dimension. The second layer includes two convolutional layers with 32 channels and a kernel size of 3. The third layer has 64 channels and a kernel size of 3, and uses max pooling with a parameter set to 2. The fourth layer includes three convolutional layers with 64 channels and a kernel size of 3.
[0033] Furthermore, the sizes of the two fully connected layers are 1024 and 256, respectively.
[0034] Furthermore, the end-to-end neural network model uses LeakyRelu as the activation function.
[0035] Compared with the prior art, the beneficial effects of the present invention are:
[0036] 1. The sonar echo simulation device provided by the present invention can autonomously complete high-resolution real-time angle of arrival estimation and generate directional simulated echoes based on array structure, which solves the problem that current echo devices cannot obtain real-time angles and improves the realism of echo simulation.
[0037] 2. This invention treats underwater acoustic angle of arrival estimation as a regression problem, directly estimating the phase weights of array elements, reducing the angle conversion steps, and making it more suitable for sonar echo simulation applications. It solves the problems of low real-time performance of traditional angle of arrival estimation methods and the fact that existing classification-based angle of arrival estimation neural networks can only classify angles as integers.
[0038] 3. Based on the excellent feature extraction and data learning capabilities of convolutional neural networks, this invention proposes a high-resolution end-to-end neural network model for underwater acoustic array signals. It does not require pre-extraction of features. Different convolutional blocks are used in the two branches to learn the real and imaginary parts of the noisy signal respectively and perform feature concatenation. Finally, the phase weight vector of the entire array is output. Attached Figure Description
[0039] Figure 1 This is a schematic diagram of the device in this invention.
[0040] Figure 2 This is a flowchart illustrating the implementation of the device in this invention.
[0041] Figure 3 This is a schematic diagram illustrating the parameter settings of the end-to-end neural network model in this invention. Detailed Implementation
[0042] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.
[0043] The specific implementation of the present invention will be described in detail below with reference to specific embodiments.
[0044] like Figure 1 As shown, a sonar echo simulation device based on array phase weight estimation is provided in one embodiment of the present invention, comprising:
[0045] The simulated target information preset module is used to preset target parameters (set the target size, distance, movement speed, etc. to be simulated) and obtain the transfer function;
[0046] The signal acquisition module is used to receive array signals;
[0047] The array signal filtering module is used to reduce noise in the array signal (this stage is the preprocessing stage, where digital filters are designed to filter out environmental noise and enhance the usability of information);
[0048] The sonar signal azimuth estimation module is used to estimate the azimuth of the sonar signal.
[0049] The echo generation module is used to reverse process the estimation results (phase weights of each array element) to generate the echo of the simulated target;
[0050] The signal transmission module is used to send the echo of the simulated target in the direction in which the sound wave needs to be transmitted (i.e., the sonar direction).
[0051] In a preferred embodiment of the present invention, the specific operation of the sonar signal azimuth estimation module in estimating the sonar signal azimuth is as follows:
[0052] The received signals of each array element are obtained, and the arrival angle direction is estimated based on the beamforming principle and convolutional neural network model. The estimation result is the phase weight of each array element.
[0053] An embodiment of the present invention provides a sonar echo simulation method based on array phase weight estimation, comprising the following steps:
[0054] Step S1: Preset target parameters and obtain the transfer function: When an underwater incident sound wave acts on a target, it will generate scattered sound waves. According to the geometric bright spot model, the scattered echo obtained by the target is regarded as the superposition of several sub-echoes, and the sub-echoes are regarded as emitted by the bright spot. The echoes generated by the bright spot are different under different incident angles of the sound waves. To simplify the calculation, the target's response to the incident wave is regarded as a linear time-invariant system, and the transfer function is expressed as:
[0055] H(r,ω)=A(r,ω)e jωτ e jφ ;
[0056] Where r is the distance between the target and the echo receiver, ω is the angular frequency, j is the imaginary unit, τ is the time delay, φ is the phase jump factor, and A(r,ω) is the amplitude factor, which is related to the target size and signal frequency. When the target echo consists of scattered sound waves emitted from multiple bright spots, the transfer functions can be linearly superimposed. Based on the bright spot method, the amplitude, time delay, and phase parameters of multiple bright spots are set to obtain the transfer function. Given the azimuth information, the scattered echoes of the target at different angles can be calculated.
[0057] Step S2: Receive array signals, reduce noise in the array signals, and estimate the sonar signal location. Beamforming is an array signal processing technique that can weighted combine signals acquired by multiple array elements to complete directional transmission and reception of the array. The array has m elements with an element spacing of d, and the sound wave is incident from an angle θ. For a uniform linear array, the time delay difference between adjacent elements is:
[0058]
[0059] Where c is the underwater acoustic speed; the frequency domain phase shift model of the signal received by the array element is as follows:
[0060]
[0061] This formula can be further transformed into:
[0062]
[0063] Assume the sound source signal vector is s n (t), where the number of narrowband sound sources is N, n m (k) is the noise vector received by the array element, then the signal collected by the linear array is:
[0064]
[0065] Based on the above mechanism, the estimation of the direction of arrival of the array received signal is transformed into the estimation of the phase weight (i.e., coefficients containing azimuth information) between each adjacent array element.
[0066] Step S3: Construct an end-to-end neural network model, use CNN to extract features of amplitude and phase and learn data, and feed the real and imaginary parts of the input signal into the end-to-end neural network model for learning respectively;
[0067] Step S4: After obtaining the phase weight results of the array elements through the end-to-end neural network model, the simulated echo is calculated based on the azimuth information input by the transfer function. The obtained phase weight information is then sent back into the array transmitting circuit to realize the end-to-end reception and simulated transmission of sonar echoes.
[0068] In this embodiment of the invention, preferably, the end-to-end neural network model uses CNN (Convolutional Neural Networks) for amplitude and phase feature extraction and data learning.
[0069] CNNs are currently widely used in fields such as computer vision and audio signal processing. A traditional CNN consists of convolutional layers, pooling layers, and fully connected layers. Convolutional layers, composed of a set of filters, perform convolution operations on the input data to extract features. Pooling layers are essentially a downsampling process used to reduce model size, improve computational speed, and enhance the robustness of the extracted features. Fully connected layers transform the input feature matrix into a one-dimensional feature vector, performing dimensionality reduction on the data. The continuous multiple layers of convolutional and pooling layers in CNNs enable them to automatically extract features. Some research converts the array's received signal into an autocovariance matrix; while this method improves computational speed, it discards many features. Directly using a similar CNN network architecture and autocovariance for fitting tasks results in poor accuracy and is unsuitable for sonar echo simulator applications. The end-to-end neural network model proposed in this invention utilizes the excellent feature extraction capabilities of CNNs, eliminating the need for separate feature extraction. In sonar echo simulation applications, DOA estimation is more suitable as a regression problem. The input signal is the array received signal. In order to learn the signal amplitude and phase characteristics more fully, the present invention separates the real part and the imaginary part and feeds them into the convolutional neural network for learning.
[0070] like Figure 3As shown, in a preferred embodiment of the present invention, the specific operation of feeding the real and imaginary parts of the input signal into the end-to-end neural network model for learning in step S3 is as follows:
[0071] When the number of array elements M = 16 and the number of snapshots is n, the input signal consists of a real matrix and an imaginary matrix of size M*n with 1 channel. The input signal is fed into a convolutional layer for feature learning. The convolutional layer includes two branches, each with four layers. The first layer includes 32 convolutional kernels, each with a kernel size of 3, using max pooling to reduce the data dimension, with a kernel size of 1 and a stride of (1,2). The second layer includes two convolutional layers with 32 channels each, and a kernel size of 3. The third layer has 64 channels, a kernel size of 3, and uses max pooling with a parameter of 2. The fourth layer includes three convolutional layers with 64 channels each, and a kernel size of 3. After the convolution operations are completed on the real and imaginary matrices, dimensionality reduction and feature concatenation are performed, and the results are fed into two fully connected layers to form a one-dimensional feature vector.
[0072] In a preferred embodiment of the present invention, the sizes of the two fully connected layers are 1024 and 256, respectively. The size of the output layer is determined by the number of array elements.
[0073] In a preferred embodiment of the present invention, the end-to-end neural network model uses LeakyRelu as the activation function.
[0074] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of the present invention, and these should also be considered within the scope of protection of the present invention. These modifications and improvements will not affect the effectiveness of the implementation of the present invention or the practicality of the patent.
Claims
1. A sonar echo simulation device based on array phase weight estimation, characterized in that, include: The simulated target information preset module is used to preset target parameters and obtain the transfer function; The signal acquisition module is used to receive array signals; The array signal filtering module is used to reduce noise in the array signal; The sonar signal azimuth estimation module is used to estimate the azimuth of the sonar signal. The specific operation of the sonar signal azimuth estimation module to estimate the sonar signal azimuth is as follows: acquire the received signals of each array element of the array structure, estimate the direction of arrival angle based on the beamforming principle and the convolutional neural network model, and the estimation result is the phase weight of each array element. The echo generation module is used to perform reverse processing based on the estimation results to generate the echo of the simulated target; The signal transmission module is used to send the echo of the simulated target in the direction of the sonar; After obtaining the phase weights of the array elements through an end-to-end neural network model, the phase weights are input according to the transfer function to calculate the simulated echo. The obtained phase weight information is then fed back into the array transmitting circuit to realize the end-to-end reception and simulated transmission of sonar echoes.
2. A sonar echo simulation method based on array phase weight estimation, characterized in that, Includes the following steps: Step S1: Preset target parameters and obtain the transfer function: Treat the scattered echo from the target as a superposition of several sub-echoes, and consider the sub-echoes as emitted from bright spots; express the transfer function as: ； Where r is the distance between the target and the echo receiving device. Here, ω is the angular frequency, and j is the imaginary unit. For time delay, It is the phase jump factor. Amplitude factor; Step S2: Receive array signals, reduce noise in the array signals, and estimate the sonar signal location; the array has m elements, the element spacing is d, and the sound wave originates from... Angular incidence; for a uniform linear array, the time delay difference between adjacent elements is: ； Where c is the underwater acoustic speed; the frequency domain phase shift model of the signal received by the array element is as follows: ； Further transformed into: ； Assume the sound source signal vector is The number of narrowband sound sources is N. Let be the noise vector received by the array element. Then the signal collected by the linear array is: ； The direction of arrival estimation of the array received signal is transformed into phase weight estimation between each adjacent array element; Step S3: Construct an end-to-end neural network model, use CNN to extract features of amplitude and phase and learn data, and feed the real and imaginary parts of the input signal into the end-to-end neural network model for learning respectively; Step S4: After obtaining the phase weight results of the array elements through the end-to-end neural network model, the phase weights are input according to the transfer function to calculate the simulated echo. The obtained phase weight information is then sent back into the array transmitting circuit to realize the end-to-end reception and simulated transmission of sonar echoes.
3. The sonar echo simulation method based on array phase weight estimation according to claim 2, characterized in that, In step S3, the specific operation of feeding the real and imaginary parts of the input signal into the end-to-end neural network model for learning is as follows: When the number of array elements M=16 and the number of snapshots is n, the input signal is a real matrix and an imaginary matrix with a size of M*n and a channel number of 1. The input signal is sent to the convolutional layer for feature learning. After the real matrix and the imaginary matrix have completed the convolution operation, dimensionality reduction and feature concatenation are performed respectively, and then sent to two fully connected layers to form a one-dimensional feature vector.
4. The sonar echo simulation method based on array phase weight estimation according to claim 3, characterized in that, The convolutional layer comprises two branches, each branch consisting of four layers. The first layer comprises 32 convolutional kernels, each with a kernel size of 3, and uses max pooling to reduce the data dimensionality. The second layer comprises two convolutional layers with 32 channels each, and a kernel size of 3. The third layer comprises 64 channels, with a kernel size of 3, and uses max pooling with a parameter set to 2. The fourth layer comprises three convolutional layers with 64 channels each, and a kernel size of 3.
5. The sonar echo simulation method based on array phase weight estimation according to claim 3, characterized in that, The sizes of the two fully connected layers are 1024 and 256, respectively.
6. The sonar echo simulation method based on array phase weight estimation according to claim 2, characterized in that, The end-to-end neural network model uses LeakyRelu as the activation function.