A method for simulating underwater acoustic signals based on WaveGAN

Abstract

The application provides a kind of underwater acoustic signal simulation method based on WaveGAN.The method comprises: using WaveGAN as underwater acoustic signal simulation model and training;Wherein, the training process specifically comprises: random noise is input to generator, and simulation underwater acoustic signal is generated by learning the distribution of real underwater acoustic signal;The simulation underwater acoustic signal and the real underwater acoustic signal are input to discriminator, and the loss function is used to make WaveGAN model converge, and the converged WaveGAN model is the underwater acoustic signal simulation model;Test sample is input to the trained generator, and simulation underwater acoustic signal is obtained.The application innovatively introduces generative adversarial network into the field of underwater acoustic signal simulation, and evaluates the simulation effect from the aspects of auditory perception, waveform, spectrogram, probability density distribution and the like through experiments, and ideal effects are obtained.

Description

Technical Field

[0001] This invention relates to the field of underwater acoustic communication technology, and in particular to a method for spoofing underwater acoustic signals based on WaveGAN. Background Technology

[0002] Since the beginning of the 20th century, the focus of global competition has gradually shifted from land to sea, leading to a surge in the development of underwater acoustic countermeasures systems. The widespread application of sonar and other underwater detection equipment, along with underwater acoustic-guided weapons, has significantly improved the effectiveness of reconnaissance and strike capabilities against enemy ships, posing a substantial threat to their survivability. Deceptive jamming is a tactic in underwater acoustic countermeasures, using equipment to mimic the acoustic characteristics of friendly ships, creating false targets to confuse enemy sonar and other underwater detection equipment, as well as underwater acoustic-guided weapons. The success of deceptive jamming hinges directly on the skill in using underwater acoustic signal mimicry techniques.

[0003] After decades of development, underwater acoustic signal imitation technology has achieved many research results. Existing underwater acoustic signal imitation methods can be divided into the following two types: (1) Based on a large number of underwater acoustic signals, empirical formulas are derived, and then the empirical formulas are used to impersonate broadband continuous spectrum and narrowband line spectrum respectively, and finally impersonated underwater acoustic signals are obtained. However, this method is difficult to adapt to the constantly changing characteristics of underwater acoustic signals. (2) Filters with specific frequency responses are used to impersonate underwater acoustic signals. However, the design of filters in this method is relatively complex, and the degree of fitting between the impersonated underwater acoustic signals and real underwater acoustic signals is low. Both of the above methods have certain limitations. Summary of the Invention

[0004] To address the aforementioned two problems in existing technologies, this invention proposes a WaveGAN-based underwater acoustic signal spoofing method. A generative adversarial network model, WaveGAN, suitable for underwater acoustic signal spoofing is constructed. After preprocessing the real underwater acoustic signal dataset, the model is used to train and optimize the parameters. The model is then used to generate spoofed underwater acoustic signals, achieving good spoofing results.

[0005] This invention provides a method for spoofing underwater acoustic signals based on WaveGAN, comprising:

[0006] Step 1: Use WaveGAN as the underwater acoustic signal spoofing model and train it; the training process specifically includes: inputting random noise into the generator, generating spoofed underwater acoustic signals by learning the distribution of real underwater acoustic signals; inputting the spoofed underwater acoustic signals and the real underwater acoustic signals into the discriminator, and using a loss function to make the WaveGAN model converge, and the converged WaveGAN model is the underwater acoustic signal spoofing model;

[0007] Step 2: Input the test sample into the trained generator to obtain the simulated underwater sound signal.

[0008] Furthermore, the function shown in formula (1) is used as the loss function:

[0009]

[0010] Where D(·) represents the discriminator, G(·) represents the generator, z represents random noise, and P z Let P represent the probability distribution of the noise, x represent the actual underwater acoustic signal, and P represent the probability distribution of the noise. data Describe the probability distribution of x. Indicates the imitation of underwater acoustic signals. express The probability distribution is given by λ, where λ represents the gradient penalty coefficient and E represents the expectation.

[0011] Furthermore, before inputting the real underwater acoustic signal into the discriminator, the real underwater acoustic signal is preprocessed; the preprocessing includes: resampling, interference suppression, and pre-emphasis.

[0012] The beneficial effects of this invention are:

[0013] This invention provides a method for spoofing underwater acoustic signals based on WaveGAN. The WaveGAN generative adversarial network model is used as the spoofing model. Real underwater acoustic signal datasets are preprocessed and used to train the model and optimize its parameters. The model is then used to generate spoofed underwater acoustic signals. Experiments were conducted to evaluate the spoofing effect from the perspectives of auditory perception, waveform, spectrogram, and probability density distribution, all of which yielded satisfactory results. Attached Figure Description

[0014] Figure 1 A flowchart illustrating a method for spoofing underwater acoustic signals based on WaveGAN, provided for an embodiment of the present invention;

[0015] Figure 2 A schematic diagram of the training process of WaveGAN provided in an embodiment of the present invention;

[0016] Figure 3 Comparison of the probability density distribution of radiated noise from counterfeit cargo ships and real cargo ships. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of the embodiments of this invention will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0018] like Figure 1 As shown, this embodiment of the invention provides a method for spoofing underwater acoustic signals based on WaveGAN, including the following steps:

[0019] S101: WaveGAN is used as the underwater acoustic signal spoofing model and trained; the training process specifically includes: inputting random noise into the generator, generating spoofed underwater acoustic signals by learning the distribution of real underwater acoustic signals; inputting the spoofed underwater acoustic signals and the real underwater acoustic signals into the discriminator, and using a loss function to make the WaveGAN model converge, and the converged WaveGAN model is the underwater acoustic signal spoofing model;

[0020] Specifically, in order to address the issues of varying absorption coefficients of seawater for different frequencies of underwater acoustic signals, significant noise interference, substantial influence of transmission distance on received signal amplitude, and non-stationary received signals, the generated simulated underwater acoustic signals should be preprocessed during training before inputting them into the discriminator. This preprocessing includes operations such as resampling, interference suppression, and pre-emphasis.

[0021] As shown in Table 1, the generator consists of six layers. The first layer comprises a fully connected layer, a reshape layer, and a ReLU activation function. The fully connected layer samples random noise. The second to fifth layers each consist of a one-dimensional deconvolution layer and a ReLU activation function. The final layer comprises a one-dimensional deconvolution layer and a Tanh activation function. The deconvolution layer generates a simulated underwater acoustic signal by learning the data distribution of real underwater acoustic signals. One-dimensional random noise of length 100 following a standard normal distribution is mapped from low-dimensional features to high-dimensional output sample data by the generator, generating a simulated underwater acoustic signal of size 16384×1.

[0022] Let the parameters of the convolution kernel be (L, D, Q), where L is the length of the convolution kernel. In the generator, the length of the convolution kernel in the Dense layer is 1, and the length of the convolution kernel in the deconvolution layer is 25. When the length of the convolution kernel is greater than 1, it can improve the receptive field. D is the depth of the convolution kernel, and its depth is equal to the number of channels of the image. Q is the number of convolution kernels, and the number of feature maps is related to the number of convolution kernels.

[0023] Table 1

[0024]

[0025]

[0026] As shown in Table 2, the discriminator consists of six layers. The first to fourth layers each consist of a one-dimensional convolutional layer, a Leaky ReLU activation function, and a phase perturbation layer. The fifth layer consists of a one-dimensional convolutional layer and a Leaky ReLU activation function, where the one-dimensional convolutional layer is used to extract feature information from the real underwater acoustic signal. The sixth layer consists of a reshape layer and a fully connected layer, where the fully connected layer is used to discriminate probabilities. The structures of convolutions and deconvolutions in the discriminator and generator, including the number of layers, kernel length, depth, and number of layers, are symmetrical.

[0027] After real and fake underwater acoustic signals are input into the discriminator, features are extracted through multi-layer convolution operations. The extracted underwater acoustic signal data features are then input into a fully connected layer for classification processing. After passing through the Sigmoid activation function, the probability of the underwater acoustic signal being real or fake is output.

[0028] Table 2

[0029]

[0030] As one possible implementation method, this embodiment takes the underwater acoustic signal of ship radiated noise as an example, such as... Figure 2 As shown, the training process is as follows: The WaveGAN model first loads preprocessed real ship radiated noise data and random noise, and inputs the random noise into the generator to obtain simulated ship radiated noise. Then, the real ship radiated noise and the generated simulated ship radiated noise are input into the discriminator. The calculated values ​​of the discriminator and generator loss functions are used to update the network weights through the Adam optimizer. The learning rate is set to 0.0001, and the exponential decay rate β1 in the momentum parameters is set to 0.5, and β2 is set to 0.9. If the preset number of training rounds is reached, the training ends; otherwise, the network training continues.

[0031] GAN models can learn the distribution of real data; however, in practice, training often presents difficulties. Therefore, the loss function needs to be redesigned. As an implementation method, this embodiment of the invention uses the Wasserstein distance with an added gradient penalty term as the loss function during training, as shown in formula (1):

[0032]

[0033] Where D(·) represents the discriminator, G(·) represents the generator, z represents random noise, and P z Let P represent the probability distribution of the noise, x represent the actual underwater acoustic signal, and P represent the probability distribution of the noise. data Describe the probability distribution of x. Indicates the imitation of underwater acoustic signals. express The probability distribution is given by λ, where λ represents the gradient penalty coefficient and E represents the expectation.

[0034] This loss function improves the stability of model training, generates higher-quality samples while ensuring sample diversity, and alleviates the overfitting problem.

[0035] S102: Input the test sample into the trained generator to obtain the simulated underwater acoustic signal.

[0036] This invention presents a wave GAN-based underwater acoustic signal spoofing method, innovatively introducing generative adversarial networks into the field of underwater acoustic signal spoofing. Experiments were conducted to evaluate the spoofing effect from the perspectives of auditory perception, waveform, spectrogram, and probability density distribution, all achieving relatively ideal results. Specifically, the auditory perception recognition rate reached 76%; the spoofed ship radiated noise is a regular envelope signal, similar in waveform to the real ship radiated noise; the energy of both the real and spoofed radiated noise signals is mainly concentrated in the low-frequency band, with very low energy in other frequency bands, exhibiting similar spectral characteristics. Figure 3 As shown, the probability density distributions of the simulated ship radiated noise signal and the real ship radiated noise signal almost overlap, exhibiting high similarity and good agreement. The experimental results verify the effectiveness of the proposed method in simulating underwater acoustic signals.

[0037] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for spoofing underwater acoustic signals based on WaveGAN, characterized in that, include: Step 1: Use WaveGAN as the underwater acoustic signal spoofing model and train it; the training process specifically includes: inputting random noise into the generator, generating spoofed underwater acoustic signals by learning the distribution of real underwater acoustic signals; inputting the spoofed underwater acoustic signals and the real underwater acoustic signals into the discriminator, and using a loss function to make the WaveGAN model converge, and the converged WaveGAN model is the underwater acoustic signal spoofing model; The generator consists of six layers. The first layer is composed of a fully connected layer, a reshape layer, and a ReLU activation function. The second to fifth layers are all composed of a one-dimensional deconvolution layer and a ReLU activation function. The last layer is composed of a one-dimensional deconvolution layer and a Tanh activation function. The discriminator consists of six layers. The first to fourth layers each include a one-dimensional convolutional layer. After each convolutional operation, a Leaky ReLU activation function and a phase perturbation layer are connected in sequence. The fifth layer includes a one-dimensional convolutional layer and a Leaky ReLU activation function in sequence. The one-dimensional convolutional layer of the fifth layer is connected to the phase perturbation layer of the fourth layer. The sixth layer includes a reshape layer and a fully connected layer in sequence. The reshape layer of the sixth layer is connected to the Leaky ReLU activation function of the fifth layer. Step 2: Input the test sample into the trained generator to obtain the simulated underwater sound signal.

2. The method for spoofing underwater acoustic signals based on WaveGAN according to claim 1, characterized in that, The function shown in formula (1) is used as the loss function: Where D(·) represents the discriminator, G(·) represents the generator, z represents random noise, and P z Let P represent the probability distribution of the noise, x represent the actual underwater acoustic signal, and P represent the probability distribution of the noise. data Describe the probability distribution of x. Indicates the imitation of underwater acoustic signals. express The probability distribution is given by λ, where λ represents the gradient penalty coefficient and E represents the expectation.

3. The method for spoofing underwater acoustic signals based on WaveGAN according to claim 1, characterized in that, Before inputting the real underwater acoustic signal into the discriminator, the real underwater acoustic signal is preprocessed; The preprocessing includes: resampling, interference suppression, and pre-emphasis.

Images