Off-grid direct wave time delay estimation method based on deep learning algorithm

Abstract

The invention discloses an off-grid direct wave time delay estimation method based on a deep learning algorithm, and relates to the field of ocean resource development, in particular to the off-grid direct wave time delay estimation method based on the deep learning algorithm. The objective of the invention is to solve the problem of poor precision of existing far-field direct wave time delay estimation. The process is as follows: an active sonar transmits signal data, and an array element receives the signal data and a time delay truth value label corresponding to the signal data; the processed signal data is obtained; obtaining a coarse estimation value of the direct wave time delay; calculating a difference value; obtaining a delay difference value after sparse processing; obtaining a trained time delay difference regression module network structure model; processing the array element receiving signal data to obtain processed signal data; outputting a feature vector based on the trained time delay difference regression module network structure model; processing the feature vector by using a peak searching function to obtain a time delay difference value; and adding the time delay difference value and the coarse estimation value of the direct wave time delay to obtain an estimation value of the received signal.

Description

Technical Field

[0001] The present invention relates to the field of marine resource development, and in particular to a method for estimating off-grid direct wave delay based on a deep learning algorithm. Background Art

[0002] Far-field direct wave delay estimation is a research topic with broad application value in underwater positioning. Accurate direct sound delay estimation is an important basis for target distance calculation, array position calibration and positioning accuracy.

[0003] In the process of ultra-short baseline positioning, an acoustic beacon is installed on the underwater target. The acoustic beacon emits an acoustic signal. After receiving the signal, the ultra-short baseline system measures the direction and distance of the target. Specifically, when the receiving device receives the signal, it records the time delay of the signal propagating to each receiving point, and calculates the location of the signal source. Existing methods include: matched filtering algorithm and model-driven and data-driven super-resolution algorithms.

[0004] The accuracy of matched filtering depends on the sampling rate of the data. When the true value of the delay is not an integer multiple of the sampling point, the matched filter will have a large error. In the scenario of high-precision direct wave delay estimation, it is necessary to seek a more accurate algorithm that is not constrained by the sampling point interval.

[0005] Model-driven methods such as the alternating direction method of multipliers (ADMM) have some parameters that need to be manually selected, such as step size parameters and penalty function parameters. Different parameter selections may lead to different convergence speeds and stability of the algorithm. In addition, some non-ideal conditions in reality, such as low signal-to-noise ratio, will cause the performance of the algorithm to deteriorate significantly.

[0006] The data-driven method is mainly a time delay algorithm based on deep learning. However, traditional deep time delay estimation uses the true value of the TOA parameter as the learning target, and as the network depth increases, the gradient will gradually disappear during the back propagation process. The overfitting problem caused by these factors will reduce the estimation accuracy of the test data. Summary of the invention

[0007] The purpose of the present invention is to solve the problem of poor accuracy of existing far-field direct wave delay estimation, and to propose an off-grid direct wave delay estimation method based on a deep learning algorithm.

[0008] A method for estimating the time delay of an out-of-grid direct wave based on a deep learning algorithm. The specific process is as follows:

[0009] Step 1: Active sonar transmits signal data, array elements receive signal data and the delay truth value label corresponding to the signal data;

[0010] Step 2: Perform data type conversion, outlier removal, modulus of real and imaginary parts of data, data normalization, etc. on the signal data received by the array element and the active sonar transmission signal data in turn to obtain processed signal data;

[0011] Step 3: Process the signal data processed in step 2 based on a matched filtering algorithm to obtain a rough estimate of the direct wave delay;

[0012] Step 4, calculating the difference between the true value of the delay corresponding to the signal data received by the array element in step 1 and the rough estimate of the direct wave delay obtained in step 3;

[0013] Step 5: Sparsely process the difference in step 4 using a sparse coding method based on a Sinc kernel function to obtain a delay difference after sparse processing;

[0014] Step 6: Use the signal data processed in step 2 as the input of the delay difference regression network structure model, use the sparsely processed delay difference obtained in step 5 as the output of the delay difference regression network structure model, train the delay difference regression module network structure model until convergence, and obtain a trained delay difference regression module network structure model;

[0015] Step 7: Process the signal data received by the array element in step 1 to obtain processed signal data;

[0016] Inputting the processed signal data into the trained network structure model of the delay difference regression module, the trained network structure model of the delay difference regression module outputs the feature vector corresponding to each processed signal;

[0017] The peak-finding function is used to process the feature vector output by the trained network structure model of the time delay regression module to obtain the time delay difference;

[0018] The delay difference is added to the rough estimate of the direct wave delay obtained in step 3 to obtain an estimate of the received signal.

[0019] The beneficial effects of the present invention are:

[0020] The present invention introduces a deep convolutional network based on sparse coding on the basis of relevant algorithms. In the delay estimation process, the coarse delay value is first calculated using the relevant processor, and then the deep convolutional network based on sparse coding is used to train and predict the difference between the coarse delay value and the actual delay value;

[0021] The present invention can break through the constraints of data sampling rate, accurately estimate the delay of non-integer multiple sampling points, and has a more accurate estimation effect when the signal noise is low;

[0022] The residual structure of the present invention introduces skip connections, which enables the network to better capture and process deeper features. At the same time, when an input is added to the function output, this operation can be clearly split into two linear superpositions, making learning easier and alleviating the training problem of deep networks.

[0023] In each algorithm module of the present invention, two convolutional layers, one pooling layer, two fully connected layers and one residual combination are used. The weight of each feature is learned by using the processing layer composed of convolutional layers, pooling layers and fully connected layers, and the normalization layer is added to the residual block to enhance the stability of the features in the network. The features are compressed by the pooling layer to obtain the global receptive field. These features are then input into two fully connected layers. Finally, the residual structure is used to avoid features that disappear during transmission between deep layers. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is the overall algorithm flow chart;

[0025] Figure 2 This is the flow chart of the data preprocessing module;

[0026] Figure 3 This is the flow chart of matched filter calculation;

[0027] Figure 4 This is the network structure diagram of the time delay regression module;

[0028] Figure 5 The change graph of training set loss when the value is 10dB;

[0029] Figure 6 It is a comparison diagram of the mean square error curve of delay estimation under different signal-to-noise ratios between the matched filtering algorithm and the present invention;

[0030] Figure 7 is the delay estimation accuracy of the present invention under different signal-to-noise ratios (confidence interval is 2ns);

[0031] Figure 8 It is a schematic diagram showing that each simulation signal data of the present invention is obtained by aliasing two signals;

[0032] Fig. 9 It is a delay grid diagram. DETAILED DESCRIPTION

[0033] Specific implementation method 1: In this implementation method, a method for estimating the time delay of a direct wave off-grid based on a deep learning algorithm is described as follows:

[0034] Step 1: Active sonar transmits signal data, array elements receive signal data and the delay truth value label corresponding to the signal data;

[0035] Step 2: Perform data type conversion, outlier removal, modulus of real and imaginary parts of data, data normalization, etc. on the signal data received by the array element and the active sonar transmission signal data in turn to obtain processed signal data;

[0036] Step 3: Process the signal data processed in step 2 based on a matched filtering algorithm to obtain a rough estimate of the direct wave delay;

[0037] Taking the two-path signal as an example, each simulated signal data is obtained by mixing two signals, and thus corresponds to two labels, such as Figure 8 ;

[0038] Assume that signal 1 (red) has a real delay difference of 1×10 -5 s, the corresponding delay grid is 1;

[0039] Assume that the actual delay difference of signal 2 (blue) is 4×10 -5 s, the corresponding delay grid is 4;

[0040] The two signals are superimposed to obtain a simulated signal, the label is (1×10 -5 ,4×10 -5 ), the corresponding grid value is (1,4); the delay grid is obtained as Fig. 9 As shown;

[0041] Active sonar transmission signal data is used to construct filters in the matched filter algorithm;

[0042] The present invention is based on the combination of a coarse delay estimation module and a delay regression module: for far-field delay estimation, if a sparse coding deep learning algorithm is used to estimate the delay, it is necessary to set grid points covering a large time domain range, so a single sparse coding method is obviously not suitable for far-field multipath delay estimation; therefore, a coarse-fine estimation process is proposed;

[0043] Delay estimation based on matched filtering achieves accurate estimation of signal delay by designing a filter that matches the expected signal and using the correlation between the signal and the filter to maximize the peak value of the signal. By observing the peak position of the matched filter output, it can be determined that the time corresponding to the peak position of the signal delay is the estimated delay of the signal.

[0044] Step 4, calculating the difference between the true value of the delay corresponding to the signal data received by the array element in step 1 and the rough estimate of the direct wave delay obtained in step 3;

[0045] Step 5: Sparsely process the difference in step 4 using a sparse coding method based on a Sinc kernel function to obtain a delay difference after sparse processing;

[0046] Step 6: Use the signal data processed in step 2 as the input of the delay difference regression network structure model, use the sparsely processed delay difference obtained in step 5 as the output of the delay difference regression network structure model, train the delay difference regression module network structure model until convergence, and obtain a trained delay difference regression module network structure model;

[0047] Step 7: Process the signal data (only data, without corresponding labels) received by the array element in step 1 to obtain processed signal data;

[0048] Inputting the processed signal data into the trained network structure model of the delay difference regression module, the trained network structure model of the delay difference regression module outputs the feature vector corresponding to each processed signal;

[0049] The peak-finding function is used to process the feature vector output by the trained network structure model of the time delay regression module to obtain the time delay difference;

[0050] 200 data items, each with 1 delay label and 1 peak value;

[0051] The delay difference is added to the rough estimate of the direct wave delay obtained in step 3 to obtain a high-precision estimate of the received signal.

[0052] The present invention can break through the constraint of time domain sampling frequency and perform high-precision time delay estimation when the time delay value is on a non-time domain sampling grid.

[0053] Specific implementation method 2: This implementation method is different from specific implementation method 1 in that the signal data received by the array element and the active sonar transmission signal data in step 2 are sequentially processed by data type conversion, outlier removal, modulus of real and imaginary parts of the data, data normalization, etc. to obtain processed signal data;

[0054] The specific process is:

[0055] Step 21: Convert the data type of the signal data received by the array element to obtain Numpy type data;

[0056] Step 22: Remove outliers from the obtained Numpy data (outliers refer to missing data);

[0057] Step 23: Take the modulus value of the real part and the imaginary part of the Numpy type data after removing the abnormal values, and obtain the data after the modulus value;

[0058] Step 24: normalize the data after taking the modulus value to obtain the processed signal data.

[0059] The other steps and parameters are the same as those in the first embodiment.

[0060] Specific implementation method three: This implementation method is different from specific implementation method one or two in that in step five, the difference in step four is sparsely processed by a sparse coding method based on a Sinc kernel function to obtain a delay difference after sparse processing;

[0061] The specific process is:

[0062] The present invention uses the Sinc function to process the function value corresponding to the delay grid to obtain a code with more sidelobe information. The steps are as follows: Figure 2 As shown;

[0063] Step 51: Set the preset delay grid size and interval;

[0064] Delay grid horizontal axis: The preset delay grid range is 0-10 (including 0 and 10), and the corresponding real time domain range is 0-10×10 -5 s (including 0 and 10), the delay grid in the range of 0-10 is divided into 1000 small grids (that is, the grid resolution (interval between adjacent grids) is set to 0.01), and the real time domain corresponding to each small grid is 10×10 -8 s(grid value);

[0065] 1000 small grids means that a delay vector is composed of 1000 0s or 1s;

[0066] Delay grid ordinate: delay value after sparse processing;

[0067] Step 52: Find the preset delay grid closest to the difference value obtained in step 4, and thin the delay difference value (such as 0.1) corresponding to the preset delay grid closest to the difference value (such as 0.11) obtained in step 4 into a set of vectors consisting of 0 and 1;

[0068] Step 53: Use Sinc to perform convolution calculation on the vector composed of 0 and 1 to obtain the convolution result p1;

[0069] Step 54: perform weighted calculation on the difference value (such as 0.11) obtained in step 4 and the out-of-grid portion (such as 0.11-0.1=0.01) of the delay difference value (such as 0.1) corresponding to the preset delay grid closest to the difference value (such as 0.11) obtained in step 4, to obtain a weighted calculation result p2;

[0070] Step 55: Add p1 and p2 to obtain the delay difference after sparse processing.

[0071] The delay difference after sparse processing is used as the training label of the deep learning input signal;

[0072] Sparse coding is used to represent the true values ​​of TOA parameters, and the network is optimized by minimizing the error between the output of the trained network and the sparse representation of the trained TOA.

[0073] The sparse coding module is based on the Sinc kernel function: after processing the real delay value corresponding to the delay grid using the Sinc function, a sparse coding with more sidelobe information can be obtained, so that the present invention has more superior accuracy and stability.

[0074] The other steps and parameters are the same as those in the first or second embodiment.

[0075] Specific implementation method 4: This implementation method is different from any one of specific implementation methods 1 to 3 in that, in step 52, the preset delay grid closest to the delay difference value after sparse processing obtained in step 3 is found, and the delay difference value (such as 0.1) corresponding to the preset delay grid closest to the delay difference value after sparse processing obtained in step 3 (such as 0.11) is sparsely processed into a set of vectors consisting of 0 and 1;

[0076] The specific process is:

[0077]

[0078] in, is the preset delay grid value (such as 0.1) closest to the delay difference after sparse processing (such as 0.11) obtained in step 3, is the preset delay grid value, δ() is the unit pulse function, D is the total number of signal paths, and i is the number of the i-th signal path.

[0079] The other steps and parameters are the same as those in Specific Embodiments 1 to 3.

[0080] Specific implementation method 5: This implementation method is different from the specific implementation methods 1 to 4 in that in step 53, Sinc is used to perform convolution calculation on the vector composed of 0 and 1 to obtain the convolution result p1;

[0081] The calculation formula is as follows:

[0082]

[0083] in, is the convolution calculation, sinc() is the sinc function.

[0084] The other steps and parameters are the same as those in Specific Embodiments 1 to 4.

[0085] Specific implementation method six: This implementation method is different from specific implementation methods one to five in that, in step five four, the delay difference value after sparse processing (such as 0.11) obtained in step three is weightedly calculated with the out-of-grid part (such as 0.11-0.1=0.01) of the delay difference value (such as 0.1) corresponding to the preset delay grid closest to the delay difference value after sparse processing (such as 0.11) obtained in step three, to obtain a weighted calculation result p2; the calculation formula is as follows:

[0086]

[0087] Among them, c is the set weighting coefficient, τ i It is the actual delay value of each path.

[0088] The other steps and parameters are the same as those in Specific Implementation Methods 1 to 5.

[0089] Specific implementation method 7: This implementation method is different from the specific implementation methods 1 to 6 in that in step 55, p1 and p2 are added to obtain the delay value after sparse processing, and the formula is as follows:

[0090] p=p1+p2

[0091] Where p is the delay value after sparse processing.

[0092] The other steps and parameters are the same as those in Specific Embodiments 1 to 6.

[0093] Specific implementation eight: This implementation differs from any one of specific implementations one to seven in that, in step six, the signal data processed in step two is used as the input of the delay difference regression network structure model, and the delay difference after sparse processing obtained in step five is used as the output of the delay difference regression network structure model, and the delay difference regression module network structure model is trained until convergence, thereby obtaining a trained delay difference regression module network structure model;

[0094] The specific process is:

[0095] The network structure model of the delay difference regression module includes the first fully connected layer FC, the first module, the second module, the third module, and the deep convolution layer;

[0096] The working process of the network structure model of the delay difference regression module is as follows:

[0097] The signal data processed in step 2 is input into the first fully connected layer FC, and the first fully connected layer FC outputs feature A;

[0098] Feature A is input into the first module, the second module, and the third module in sequence, and the third module outputs feature A′;

[0099] The output feature A′ of the third module is summed element by element with the output feature A of the first fully connected layer FC to obtain feature A″;

[0100] Feature A″ is input into the deep convolution layer, and the deep convolution layer outputs features;

[0101] The output feature of the deep convolutional layer is the delay difference after sparse processing obtained in step 5.

[0102] The other steps and parameters are the same as those in Specific Embodiments 1 to 7.

[0103] Specific implementation nine: This implementation is different from any of specific implementations one to eight in that each of the first module, the second module, and the third module includes: a first convolutional layer, a first batch normalization BN, a first ReLuctant activation function layer, a second convolutional layer, a second batch normalization BN, a second ReLuctant activation function layer, a maximum pooling layer, a second fully connected layer FC, a third ReLuctant activation function layer, a third fully connected layer FC, and a fourth ReLuctant activation function layer.

[0104] The other steps and parameters are the same as those in Specific Embodiments 1 to 8.

[0105] Specific implementation method 10: This implementation method is different from the specific implementation methods 1 to 9 in that the working process of each module in the first module, the second module and the third module is as follows:

[0106] Feature A is sequentially input into the first convolutional layer, the first batch normalization BN, the first ReLuctant activation function layer, the second convolutional layer, the second batch normalization BN, the second ReLuctant activation function layer, and the second ReLuctant activation function layer outputs feature B;

[0107] Feature B is sequentially input into the maximum pooling layer, the second fully connected layer FC, the third ReLuctant activation function layer, the third fully connected layer FC, the fourth ReLuctant activation function layer, and the fourth ReLuctant activation function layer outputs feature C;

[0108] Feature C is the output feature of each module.

[0109] The other steps and parameters are the same as those in Specific Embodiments 1 to 9.

[0110] 1) Convolution operation, by sliding a small filter on the input data, calculating the dot product of the filter and the local area of ​​the input data, and obtaining the output feature map. This operation can extract the features of the input data and is a key step for feature extraction in the CNN model.

[0111] 2) Activation functions are used to introduce nonlinearity so that the model can learn more complex features. The role of these functions is to enhance the expressiveness of the model so that it can handle a wider range of input data.

[0112] 3) Pooling layer, which is used to reduce the dimension of feature maps and the amount of computation while retaining important feature information. Commonly used pooling methods include maximum pooling and average pooling. The role of the pooling layer is to further extract and concentrate features, making the model more robust to small changes in the input data.

[0113] 4) Fully connected layer, which integrates the features extracted by the previous convolutional layer and pooling layer, performs linear transformation through weights and biases, and then performs nonlinear transformation through activation functions to finally obtain the classification result. The function of the fully connected layer is to map the learned features to the sample label space to complete the classification or regression task.

[0114] 5) Normalization layer: Normalize the output of each layer to speed up training and improve model stability. Common ones include batch normalization and layer normalization.

[0115] The delay difference regression module in the present invention introduces a residual structure: in order to reduce the risks of gradient vanishing and overfitting during the training of deep neural networks, a residual structure is introduced. The residual structure enables the network to better capture and process deeper features by introducing jump connections. As the depth of the network increases, the gradient will gradually disappear during the back propagation process, making the network difficult to train. The residual structure directly connects the input to the output through a jump connection, so that the gradient can flow directly back from the output to the input layer, thereby alleviating the problem of gradient vanishing. At the same time, by adding an input to the function output, the relationship between the input and output can still be described. This operation can be clearly split into two linear superpositions, making learning simpler, while alleviating the training problem of deep networks and providing the possibility for the training of thousands of layers of networks.

[0116] The delay difference regression module in the present invention encapsulates the deep convolutional network with residual structure into an algorithm module: each algorithm module consists of two convolutional layers, one pooling layer, two fully connected layers and a residual structure. The specific structure is as follows:

[0117] Structure 1) Use two convolutional layers to learn the weight of each feature channel and extract TOA features.

[0118] Structure 2) uses a pooling layer to compress features to obtain a global receptive field.

[0119] Structure 3) uses two fully connected layers to calculate the nonlinear relationship of the compressed feature channels, where the Relu function is used as the activation function.

[0120] Structure 4) uses residual blocks to avoid the disappearance of part of TOA features during transmission between deep layers.

[0121] The above structure is encapsulated as an algorithm module. When presetting parameters, the number of neural network layers can be quickly increased by directly changing the number of algorithm modules. At the same time, the residual structure can effectively avoid the occurrence of overfitting and improve the performance of network feature extraction.

[0122] Example:

[0123] The method for estimating the time delay of a direct wave from a grid based on a deep learning algorithm in a specific embodiment of the present invention is implemented in the following steps: Figure 1 As shown, the details are as follows:

[0124] Step 1) In the scenario of far-field direct wave delay estimation, the correlation algorithm, correlation interpolation method and the proposed deep learning delay estimation algorithm based on sparse coding are simulated. The simulation signal parameters are as follows: signal length 10ms, frequency domain range 20kHz-30kHz, sampling frequency 100kHz, number of array elements 1, signal delay value 100×10 -5 s to 150×10 -5 Any value in s.

[0125] Step 2) Set the required parameters in the algorithm module. Collect 100,000 samples in total, randomly divide them into 90% for training and 10% for validation. The background noise is a white Gaussian noise corresponding sampling signal with additive noise. The training is based on Pytorch and Adam optimizer. The true TOA values ​​of the training set and the test set are randomly generated within the specified interval. For algorithms containing sparse coding, the length of the predefined overcomplete dictionary is 1000. Calculate the root mean square error (RMSE) of all TOA estimates to evaluate the estimation accuracy.

[0126] Step 3), use the data preprocessing module to perform the following processing, and implement the steps as follows Figure 2 As shown, the details are as follows:

[0127] Step 3-1), perform type conversion on the array element receiving data and the active sonar transmitting signal data to obtain Numpy type data;

[0128] Step 3-2), encapsulate the data into a three-dimensional array with a dimension of 200×2×1200, the number of training data in each batch, where the received signal is the first row of data in the second dimension, and the transmitted signal is the second row of data in the second dimension;

[0129] Step 3-3), remove outliers from the converted type data to eliminate the interference of outliers on signal processing;

[0130] Step 3-4), take the modulus value of the real part and the imaginary part of the processed data and perform normalization processing to obtain standard data.

[0131] Step 4) Calculate the rough estimate of the delay using the delay rough estimation module. Figure 3 As shown, the details are as follows:

[0132] Step 4-1) Use the transmitted signal data in the input array to construct a matched filter for the desired signal.

[0133] Step 4-2) Correlation operation is performed on the data of the received signal in the input array and the constructed matched filter.

[0134] Step 4-3) Find the peak position in the correlated data and subtract half the value of the preset delay estimation grid length as the output value of the delay rough estimation module.

[0135] Step 4-4), delete the transmission signal data in the last input array.

[0136] Step 5) Use the sparse coding module to calculate the label of the delay estimation. The specific steps are as follows:

[0137] Step 5-1) Subtract the rough estimate value output by the delay estimation module from the actual delay value to obtain a residual value.

[0138] Step 5-2) Use the Sinc function to process the function value corresponding to the delay grid to obtain a code with more sidelobe information. The off-grid value is calculated as the additional value at the nearest grid position, and the off-grid delay can also be calculated based on the additional value.

[0139] Step 6), input the sparsely coded labels and the signal data output by the delay rough estimation module into the delay regression module network structure for training, as shown in the steps Figure 4 As shown, the details are as follows:

[0140] Step 6-1) Build a deep learning network structure according to preset parameters. The network structure consists of several algorithm modules, each of which consists of two convolutional layers, one pooling layer, two fully connected layers and a residual structure. The specific structure is as follows:

[0141] Structure 1) Use two convolutional layers to learn the weight of each feature channel and extract TOA features.

[0142] Structure 2) Use a pooling layer to compress the features to obtain the global receptive field.

[0143] Structure 3) uses two fully connected layers to calculate the nonlinear relationship of the compressed feature channels, where the ReLU function is selected as the activation function.

[0144] Structure 4) Use residual blocks to avoid the disappearance of part of the TOA features during transmission between deep layers.

[0145] When presetting parameters, the number of modules is set to 3, with a total of 6 convolutional layers, 3 pooling layers, 6 fully connected layers, and 3 residual structures.

[0146] The convolution kernel size, step size, and hyper parameters of the fully connected layer in the present invention are as follows: Figure 4 as shown in .

[0147] Step 6-2), randomly set the weights and bias values ​​of the input layer, and establish a nonlinear relationship between layers during the forward propagation process. During the forward propagation process, the input data passes through the input layer, the hidden layer, and is processed layer by layer and transmitted to the output layer.

[0148] Step 6-3) Calculate the sum of squares of the output layer and the set label error value as the constraint function, use the partial derivative of the objective function to calculate the weight of each neuron in the back propagation, and use the stochastic gradient descent algorithm to gradually correct the weights.

[0149] Step 6-4), repeat steps 6-2) and 6-3) until the number of training times reaches a preset value.

[0150] Step 6-5) Figure 5 This is a graph showing the change in the neural network training set loss when the signal-to-noise ratio is set to 10dB.

[0151] Step 7), repeat steps 1) to 6) at multiple signal-to-noise ratios in the range of -5 to 10 dB and with a step length of 3 dB, and obtain the mean square error curve of the delay estimation of the present invention at different signal-to-noise ratios as shown in Figure 6 As shown, when the confidence interval is 0.25 adjacent sampling points, the delay estimation accuracy under different signal-to-noise ratios is calculated as follows: Figure 7 shown.

[0152] Step 8) Figure 6 It can be seen that when the signal-to-noise ratio is -5 to 10dB and the step size is 3dB, the RMSE value of the algorithm of the present invention is significantly reduced compared with the matched filter-based algorithm. That is, in the far-field direct wave delay estimation scenario, the delay estimation accuracy of the algorithm proposed in this paper is significantly improved compared with the related delay estimation algorithm. When SNR<10dB, the algorithm RMSE value drops by more than 66%.

[0153] Step 9), by Figure 7 It can be seen that when the signal-to-noise ratio is -5 to 10 dB, the delay estimation accuracy of the algorithm of the present invention is relatively high, all above 75%, while the delay estimation accuracy of the matched filtering algorithm is below 45%.

[0154] The present invention may also have many other embodiments. Without departing from the spirit and essence of the present invention, those skilled in the art may make various corresponding changes and modifications based on the present invention, but these corresponding changes and modifications should all fall within the scope of protection of the claims attached to the present invention.

Claims

1. A method for estimating the time delay of an out-of-grid direct wave based on a deep learning algorithm, characterized in that: The specific process of the method is: Step 1: Active sonar transmits signal data, array elements receive signal data and the delay truth value label corresponding to the signal data; Step 2: Perform data type conversion, outlier removal, modulus of real and imaginary parts of data, data normalization, etc. on the signal data received by the array element and the active sonar transmission signal data in turn to obtain processed signal data; Step 3: Process the signal data processed in step 2 based on a matched filtering algorithm to obtain a rough estimate of the direct wave delay; Step 4, calculating the difference between the true value of the delay corresponding to the signal data received by the array element in step 1 and the rough estimate of the direct wave delay obtained in step 3; Step 5: Sparsely process the difference in step 4 using a sparse coding method based on a Sinc kernel function to obtain a delay difference after sparse processing; Step 6: Use the signal data processed in step 2 as the input of the delay difference regression network structure model, use the sparsely processed delay difference obtained in step 5 as the output of the delay difference regression network structure model, train the delay difference regression module network structure model until convergence, and obtain a trained delay difference regression module network structure model; Step 7: Process the signal data received by the array element in step 1 to obtain processed signal data; Inputting the processed signal data into the trained network structure model of the delay difference regression module, the trained network structure model of the delay difference regression module outputs the feature vector corresponding to each processed signal; The peak-finding function is used to process the feature vector output by the trained network structure model of the time delay regression module to obtain the time delay difference; The delay difference is added to the rough estimate of the direct wave delay obtained in step 3 to obtain an estimate of the received signal.

2. According to claim 1, a method for estimating the time delay of an out-of-grid direct wave based on a deep learning algorithm is characterized in that: In the step 2, the signal data received by the array element and the active sonar transmission signal data are sequentially processed by data type conversion, outlier removal, modulus of real and imaginary parts of the data, data normalization, etc. to obtain processed signal data; The specific process is: Step 21: Convert the data type of the signal data received by the array element to obtain Numpy type data; Step 22: Remove outliers from the obtained Numpy data; Step 23: Take the modulus value of the real part and the imaginary part of the Numpy type data after removing the abnormal values, and obtain the data after the modulus value; Step 24: normalize the data after taking the modulus value to obtain the processed signal data.

3. The off-grid direct wave delay estimation method based on deep learning algorithm according to claim 2 is characterized by: In the step 5, the difference in step 4 is subjected to sparse processing by using a sparse coding method based on a Sinc kernel function to obtain a delay difference after sparse processing; The specific process is: Step 51: Set the preset delay grid size and interval; Step 52: Find the preset delay grid closest to the difference obtained in step 4, and thin the delay difference corresponding to the preset delay grid closest to the difference obtained in step 4 into a set of vectors consisting of 0 and 1; Step 53: Use Sinc to perform convolution calculation on the vector composed of 0 and 1 to obtain the convolution result p1; Step 54: perform weighted calculation on the difference obtained in step 4 and the out-of-grid portion of the delay difference corresponding to the preset delay grid closest to the difference obtained in step 4 to obtain a weighted calculation result p2; Step 55: Add p1 and p2 to obtain the delay difference after sparse processing.

4. The off-grid direct wave delay estimation method based on deep learning algorithm according to claim 3 is characterized by: In the step 52, find the preset delay grid closest to the delay difference after sparse processing obtained in step 3, and thin the delay difference corresponding to the preset delay grid closest to the delay difference after sparse processing obtained in step 3 into a set of vectors consisting of 0 and 1; The specific process is: in, is the preset delay grid value closest to the delay difference after sparse processing obtained in step 3, is the preset delay grid value, δ( ) is the unit pulse function, D is the total number of signal paths, and i is the number of the i-th signal path.

5. The method for estimating the time delay of an out-of-grid direct wave based on a deep learning algorithm according to claim 4, characterized in that: In the step 53, Sinc is used to perform convolution calculation on the vector composed of 0 and 1 to obtain the convolution result p1; The calculation formula is as follows: in, is the convolution calculation, sinc() is the sinc function.

6. The method for estimating the time delay of an off-grid direct wave based on a deep learning algorithm according to claim 5, characterized in that: In the step 54, the delay difference value after sparse processing obtained in step 3 and the out-of-grid portion of the delay difference value corresponding to the preset delay grid closest to the delay difference value after sparse processing obtained in step 3 are weighted calculated to obtain a weighted calculation result p2; The calculation formula is as follows: Among them, c is the set weighting coefficient, τ i It is the actual delay value of each path.

7. The method for estimating the time delay of an out-of-grid direct wave based on a deep learning algorithm according to claim 6, characterized in that: In the step 55, p1 and p2 are added to obtain the delay value after sparse processing; The formula is as follows: p=p1+p2 Where p is the delay value after sparse processing.

8. The method for estimating the time delay of an off-grid direct wave based on a deep learning algorithm according to claim 7, characterized in that: In the step 6, the signal data processed in the step 2 is used as the input of the delay difference regression network structure model, and the delay difference after sparse processing obtained in the step 5 is used as the output of the delay difference regression network structure model, and the delay difference regression module network structure model is trained until convergence, so as to obtain a trained delay difference regression module network structure model; The specific process is: The network structure model of the delay difference regression module includes the first fully connected layer FC, the first module, the second module, the third module, and the deep convolution layer; The working process of the network structure model of the delay difference regression module is as follows: The signal data processed in step 2 is input into the first fully connected layer FC, and the first fully connected layer FC outputs feature A; Feature A is input into the first module, the second module, and the third module in sequence, and the third module outputs feature A′; The output feature A′ of the third module is summed element by element with the output feature A of the first fully connected layer FC to obtain feature A″; Feature A″ is input into the deep convolution layer, and the deep convolution layer outputs features; The output feature of the deep convolutional layer is the delay difference after sparse processing obtained in step 5.

9. The method for estimating the time delay of an off-grid direct wave based on a deep learning algorithm according to claim 8, characterized in that: Each of the first module, the second module and the third module includes: a first convolutional layer, a first batch normalization BN, a first ReLuctant activation function layer, a second convolutional layer, a second batch normalization BN, a second ReLuctant activation function layer, a maximum pooling layer, a second fully connected layer FC, a third ReLuctant activation function layer, a third fully connected layer FC, and a fourth ReLuctant activation function layer.

10. The off-grid direct wave delay estimation method based on deep learning algorithm according to claim 9, characterized in that: The working process of each module in the first module, the second module and the third module is as follows: Feature A is sequentially input into the first convolutional layer, the first batch normalization BN, the first ReLuctant activation function layer, the second convolutional layer, the second batch normalization BN, the second ReLuctant activation function layer, and the second ReLuctant activation function layer outputs feature B; Feature B is sequentially input into the maximum pooling layer, the second fully connected layer FC, the third ReLuctant activation function layer, the third fully connected layer FC, the fourth ReLuctant activation function layer, and the fourth ReLuctant activation function layer outputs feature C; Feature C is the output feature of each module.

Images