Model data dual-drive-based anti-interference underwater acoustic channel estimation method and device

Abstract

The invention discloses an anti-interference underwater acoustic channel estimation method and device based on model data dual drive, and the method comprises the steps: obtaining an interference data set, and training a neural network; preprocessing the underwater acoustic signal to obtain a pilot frequency part signal, and initializing a channel, an interference sample, a sampling moment and a channel precedence variance gamma; respectively obtaining interference and channel prior scores at the t'moment by using a neural network and an analytical model; the logarithmic likelihood function obtains a gradient about interference and channel conjugation to obtain interference and channel condition scores, and the prior and condition scores are weighted and summed to obtain interference and channel posterior scores; the channel samples and the interference samples are corrected through annealing sampling, gamma is updated, and the posterior score is calculated again; a channel sample and an interference sample at the t'moment are obtained through the inverse process of the diffusion model; and if t'reaches tmin, outputting a final estimated value, otherwise, updating gamma according to the channel sample at the t 'moment, and circulating. According to the invention, accurate estimation of the single carrier communication channel in the structured interference environment is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of underwater acoustic signal processing, and in particular to a model data dual-driven anti-interference underwater acoustic channel estimation method and device. BACKGROUND

[0002] The rapid development of underwater sensing and network systems has promoted the widespread application of unmanned underwater vehicles, sensors and autonomous platforms in ocean monitoring, resource exploration and multi-machine cooperative tasks. However, these technological advances face a fundamental challenge: the limited bandwidth of underwater acoustic communication. When environmental perception, inter-platform communication and real-time control are used in the same frequency band, signal overlap between coexisting systems will be inevitable. For example, when an unmanned underwater vehicle formation is receiving communication signals from the control end, some platforms may simultaneously transmit detection signals in the same frequency band, or use underwater acoustic communication in the same frequency band for data exchange within the formation. Such interference will make the channel estimation and demodulation process extremely complex. Traditional channel estimation methods assume that the background noise is Gaussian white noise, and it is difficult to accurately estimate the channel when structured interference occurs. It is of great significance to study channel estimation methods under structured interference.

[0003] Although the number of parameters to be estimated has increased, compared with the method of regarding interference as additive Gaussian white noise, jointly estimating the channel and interference signal and effectively utilizing the prior information of interference is expected to improve performance. Although the diffusion model is originally designed to sample from the base distribution unconditionally, it has been extended to be applicable to conditional generation tasks by introducing observation data based on the likelihood framework. In communication systems, channel estimation is constructed as a linear inverse problem, whose goal is to recover the channel state from noisy observation data. Recent research has begun to use data-driven diffusion models to learn the characteristics of multiple-input multiple-output channel structures to complete the channel estimation task, further improving the accuracy of channel estimation under Gaussian white noise. However, in the single-input single-output scenario, the channel has no obvious characteristics other than the sparse structure, and it is difficult to obtain water acoustic channel samples. Therefore, using a purely data-driven diffusion model to estimate the single-input single-output channel in a structured interference environment faces problems such as insufficient data samples and redundant computational complexity.

[0004] The patent with publication number CN120074992A discloses a Bayesian wireless channel estimation method and system suitable for co-channel interference, which is based on variational inference of sub-channel link level interference (LSI) to capture the large interference power frequency points in the considered frequency band, can more effectively extract the frequency points where the real interference exists from the actual frequency selective fading channel, and perform alternating variational inference based on global interference (GI) and LSI for optimization algorithm model and reduction of algorithm overfitting. An interference level hyperparameter adaptive updating method is introduced to introduce more prior information entropy in iteration. The invention can complete position estimation by combining ToA / TDoA / AoA algorithms in the channel parameter estimation of the pilot frequency band of OFDM, can complete navigation prediction by combining Kalman / Particle Filter, and can use correlation equalization algorithm to improve the communication quality of signal subcarriers. However, the interference signal targeted by the invention is limited to the electromagnetic path commonly used in the device or the interference caused by internal circuit leakage. These interferences often exist in the form of sine waves or periodic pulses, have strong structure, and can be modeled using analytical methods, making it difficult for the method to be applicable to more complex interference signals.

[0005] The patent with publication number CN112737702A discloses a MIMO underwater acoustic channel estimation method in a sparse interference background. The invention uses variational Bayesian inference theory to estimate the pulse interference and the sparsity of the channel respectively, and fully utilizes the spatial correlation of the channel to obtain processing gain. Compared with existing underwater acoustic channel estimation methods, the invention has higher estimation accuracy and lower computational complexity. However, the invention is limited to pulse signals for interference signals, and can use sparse priors for prior probability modeling. For complex signals, there is no analytical prior probability, and they are not sparse, so the invention cannot be applied. SUMMARY

[0006] To overcome the shortcomings of the prior art, the present invention proposes a model data dual-driven anti-interference underwater acoustic channel estimation method and device, which uses analytical modeling channel priors and neural network learned interference priors to accurately estimate single carrier communication channels in a structured interference environment when complex scene single carrier communication signals are interfered by probe signals or other modulated communication signals.

[0007] The specific technical solutions are as follows: A model data dual-driven anti-interference underwater acoustic channel estimation method, comprising the following steps: S1: Obtain an interference sample data set, and train a neural network based on a U-net structure; S2: preprocessing the received underwater acoustic signal to obtain a pilot part signal y after symbol rate sampling, initializing K channel samples h and K interference samples n using Gaussian white noise, initializing sampling time t'=1, and channel prior variance γ; S3: setting sampling time t=t', obtaining interference prior score at time t using the neural network, obtaining channel prior score at time t using an analytical prior model of the channel samples, obtaining log-likelihood function at time t using an approximation method, obtaining interference conditional score at time t by taking the gradient of the log-likelihood function with respect to the conjugate of the interference, obtaining channel conditional score at time t by taking the gradient with respect to the conjugate of the channel, obtaining interference posterior score at time t by weighted sum of the prior score and the conditional score of the interference, and obtaining channel posterior score at time t by weighted sum of the prior score and the conditional score of the channel; S4: using annealed Langevin sampling method to correct the channel samples and the interference samples according to the posterior scores, and updating the channel prior variance according to the corrected channel samples; S5: using the method of S3 to obtain posterior scores corresponding to the corrected channel samples and the interference samples, obtaining channel samples and interference samples at the next time using the inverse process of the diffusion model, and updating t'=t+dt, where dt is the inverse time step; S6: determining whether t' reaches a set threshold t min , if yes, outputting a final estimation value, and if no, updating the channel prior variance according to the channel samples at the next time obtained in S5, and returning to execute S3-S6; the final estimation value includes the mean of the K channel samples at time t' and the mean of the K interference samples at time t' .

[0008] Further, in S1, the neural network comprises an encoder, a decoder, and a skip connection module; the real part and the imaginary part of the input data are respectively taken as independent channels, stacked to form a two-channel tensor, and projected to a 64-dimensional feature space through an initial convolution layer as the input of the encoder; The encoder and the decoder are both composed of 32 cascaded convolution modules, and each convolution module contains a convolution layer and a ReLU activation function; The skip connection module stacks the output feature maps of each convolution module in the encoder and the input feature maps of the corresponding level in the decoder in the channel to enhance gradient flow and preserve multi-scale features; The feature maps output by the decoder are projected to a two-dimensional space through a final convolution layer to generate interference prior score estimation results.

[0009] Further, in S3, the real part of the interference prior score is the jth interference sample n The first column of the output after inputting into the neural network; the imaginary part is... The second column of the output after inputting into the neural network; Channel prior score at time t The gradient is obtained by multiplying the channel prior distribution with the channel perturbation kernel function at time t and then taking the gradient with respect to the channel conjugate at time t.

[0010] Furthermore, the method of obtaining the log-likelihood function at time t using approximation means specifically involves: representing the interference in the pilot signal reception model by scaling the interference at time t and adding noise, and representing the channel by scaling the channel at time t and adding noise, thereby obtaining the log-likelihood function at time t.

[0011] Furthermore, S4 is specifically implemented through the following sub-steps: (4.1) Calculate the channel sample correction step size: The channel correction step size is a hyperparameter divided by the square of the L2 norm of the channel fraction; Obtain the corrected channel sample It is the sum of the product of the original channel sample, the channel correction step size and the channel posterior score, the square root of twice the channel correction step size and the standard Gaussian white noise; (4.2) Calculate the interference sample correction step size: The channel correction step size is a hyperparameter divided by the square of the L2 norm of the interference fraction; Obtain the corrected interference samples The sum of the original interference sample, the interference correction step size and the interference posterior score, the square root of twice the interference correction step size and the standard Gaussian white noise; (4.3) Use the expectation-maximization algorithm to update the channel prior variance based on the corrected channel samples.

[0012] Further, in S5, the channel sample at the next time step is obtained by using the inverse process of the diffusion model. Specifically, the channel sample at the next time step is obtained by subtracting the second term from the corrected channel sample at the current time step and adding the third term. The second term is the sum of half the product of the drift coefficient of the diffusion process and the corrected channel sample at the current time step, the product of the drift coefficient and the channel posterior score, and the product of the inverse time step. The third term is the product of the square root of the negative drift coefficient and the inverse time step step and the standard Gaussian white noise. The next time interference sample is obtained by using the diffusion model inverse process, specifically: the next time interference sample is obtained by subtracting a new second term from the current time corrected interference sample as a new first term, adding a new third term, the new second term is the product of the diffusion process drift coefficient and the current time corrected interference sample, half of the sum of the product of the drift coefficient and the interference posterior score, and the product of the inverse time step; the new third term is the product of the negative drift coefficient and the inverse time step after taking the square root, and the product of the standard Gaussian white noise.

[0013] A model data double-driven anti-interference underwater acoustic channel estimation device, comprising a memory and one or more processors, the memory stores executable code, and the one or more processors execute the executable code to implement the model data double-driven anti-interference underwater acoustic channel estimation method.

[0014] A computer readable storage medium having a program stored thereon, the program being executed by a processor to implement the model data double-driven anti-interference underwater acoustic channel estimation method.

[0015] The beneficial effects of the present application are: The present application solves the problem that single carrier communication channel cannot be accurately estimated when the same frequency band and same time single carrier communication signal is interfered by a detected signal or other modulated communication signal, and realizes accurate estimation of single carrier communication channel in a structured interference environment. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 It is a flowchart of the model data double-driven anti-interference underwater acoustic channel estimation method in the embodiment of the present application.

[0017] Figure 2 It is a structural schematic diagram of a neural network in the embodiment of the present application.

[0018] Figure 3 It is a curve diagram of the change of channel sound velocity with depth used in simulation in the embodiment of the present application.

[0019] Figure 4 It is a channel estimation NMSE performance diagram when the interference signal is a parameter random HFM or LFM signal in the embodiment of the present application.

[0020] Figure 5 It is a channel estimation NMSE performance diagram of the model data double-driven anti-interference underwater acoustic channel estimation method in the embodiment of the present application under the transmission of random OFDM interference.

[0021] Figure 6 It is a structural schematic diagram of the model data double-driven anti-interference underwater acoustic channel estimation device in the embodiment of the present application. DETAILED DESCRIPTION

[0022] The present application will be described in detail below with reference to the accompanying drawings and preferred embodiments, the objects and effects of the present application will become more apparent. It should be understood that the specific embodiments described herein are merely intended for the purpose of illustration of the present application and are not intended to limit the present application.

[0023] As shown in the figure, a model data dual-driven anti-interference underwater acoustic channel estimation method includes the following steps: Figure 1 S1: The analytical prior is used to model the prior and its score (i.e. the gradient of the log-likelihood function) of the channel, and the analytical prior model of the channel to be estimated h is:

[0024] In the formula, ( ) represents a complex Gaussian distribution function; γ represents the channel variance, which is used to control the sparsity.

[0025] Obtain the interference sample data set, build a neural network as shown in the figure Figure 2 The neural network adopts a structure similar to U-Net and has a full convolution layer; the neural network includes an encoder, a decoder, and a skip connection module. The real and imaginary parts of the input data are stacked to form a two-channel tensor as independent channels, and are projected to 64 feature dimensions through an initial convolution layer as the input of the encoder. The encoder and the decoder each contain 32 cascaded convolution modules, each of which contains a convolution layer and a ReLU activation function; the skip connection module stacks the output feature maps of each convolution module in the encoder with the input feature maps of the corresponding level in the decoder, thereby enhancing gradient flow and preserving feature integrity. The feature maps output by the decoder are projected to a two-dimensional space through a final convolution layer to generate estimated interference prior scores.

[0026] The parameters of the neural network are denoted as θ, and the initialization hyperparameters β min and β max are obtained by training the neural network according to the interference sample data set. The loss function J(θ) of the neural network is expressed as follows:

[0027]

[0028]

[0029]

[0030]

[0031] ​In the formula, n is the interference sample, n t is the interference sample after noise disturbance at time t, n t * represents the conjugate of n t represents the gradient operator of n t * represents the disturbance kernel function; p(t) represents the distribution function of time, which is a uniform distribution function between 0 and 1; p data (n) is the empirical distribution of the interference sample data set; is a neural network with input and output dimensions of Mx2 as shown in Figure 2 , M is the length of n t ; a(t) is a scrambling parameter, I M represents an MxM identity matrix, β(t) is a drift coefficient, β min and β max are hyperparameters for controlling the relationship between noise variance and time in the disturbance kernel function.

[0032] In this embodiment, the specific implementation is: randomly selecting B groups of interference samples from the interference sample data set, adding noise to each group of interference samples using the disturbance kernel function, and separating the real and imaginary parts of the noise-disturbed samples to form a BxMx2-dimensional tensor input into the neural network to obtain the loss function, and optimizing the parameters θ using the adaptive matrix estimation (Adma) optimizer according to the loss function, and saving the network parameters θ after training is completed.

[0033] S2: Preprocess the received underwater acoustic signal to obtain the pilot part signal y after symbol rate sampling, initialize the sampling samples of K to-be-estimated variables (including channel samples and interference samples) using Gaussian white noise, initialize the variance of the Gaussian white noise , a hyperparameter t min for controlling the step size at the sampling stop time (i.e., setting a threshold), initialize t'=1, the channel prior variance at this time is γ=1 L , in this embodiment, γ=1 L , that is, a column vector of 1.

[0034] The expression of the pilot part signal is as follows:

[0035] In the formula, is the received pilot part baseband data, is a toplex matrix composed of the pilot after scaling by 1 / L, L is the channel length; is the to-be-estimated channel, is Gaussian white noise. ​​

[0036] The first column elements of the Toeplitz matrix A are defined as:

[0037] where N is the pilot length, and N-L+1=M.

[0038] The first row elements of the Toeplitz matrix A are defined as:

[0039] K channel samples are initialized using Gaussian white noise, expressed as follows:

[0040] where I L is an LxL identity matrix, i=1,2,3,…,K.

[0041] K interference samples are initialized using Gaussian white noise, expressed as follows:

[0042] where j=1,2,3,…,K.

[0043] In this embodiment, a SISO underwater acoustic channel is generated by Bellhop acoustic simulation software, with the following parameters: water depth of 5000m, distance of 10km, transmitter depth of 600m to 800m random number each time, receiver depth of 1000-1200m random number, and sound velocity curve varying with depth as shown in Figure 3

[0044] S3: Let the sampling time t=t', obtain the interference prior score at time t using the neural network obtained by S1, calculate the channel prior score at time t using the analytical prior model of the channel, obtain the interference conditional score at time t by taking the gradient of the likelihood function at time t with respect to the conjugate of the interference, obtain the channel conditional score at time t by taking the gradient of the likelihood function at time t with respect to the conjugate of the channel, obtain the interference posterior score at time t by weighted sum of the prior score and the conditional score of the interference, and obtain the channel posterior score at time t by weighted sum of the prior score and the conditional score of the channel; the specific implementation is realized through the following sub-steps: (3.1) Calculate the channel prior score at time t using the analytical prior model of the channel, expressed as follows:

[0045]

[0046] where is the i-th channel sample at time t, represents the conjugate of h t ,​ express The gradient operator.

[0047] The prior disturbance score at time t is obtained using a neural network, expressed as follows:

[0048] In the formula, Let j be the j-th interference sample at time t. express The gradient operator; Let t be the first column of the output result after the j-th interference sample is input into the neural network, representing the real part of the prior score; The second column of the output after the j-th interference sample at time t is input into the neural network represents the imaginary part of the prior score.

[0049] (3.2) Obtain the scrambled log-likelihood function at time t using approximation methods. The approximation methods include: representing the interference in the pilot signal reception model by scaling it up to the interference at time t and adding noise, and representing the channel by scaling it up to the channel at time t and adding noise, thus obtaining the log-likelihood function at time t. Then, the channel conditional score is obtained by calculating the gradient of the log-likelihood function with respect to the channel conjugate. The conditional score of the disturbance is obtained by taking the gradient of the log-likelihood function with respect to the conjugate of the disturbance. The expression is as follows:

[0050] in

[0051] (3.3) The channel posterior score at time t is obtained by weighted summation of the prior score and conditional score of the channel, as expressed below:

[0052] In the formula, Indicates all The set, Indicates all The set of values, where μ is a hyperparameter used to control the weights of the channel prior scores, is set manually.

[0053] The posterior score of the disturbance at time t is obtained by weighted summation of the prior score and conditional score, as shown in the following expression:

[0054] In the formula, This is a hyperparameter used to control the weight of the perturbation prior scores; it is set manually.

[0055] S4: According to the result of S3, the channel samples and interference samples are corrected one by one using the annealed Langevin dynamics method, and the channel variance is updated. Annealed Langevin Dynamics is a sampling method that combines Langevin dynamics and annealing strategies, widely used in generative models (such as image generation, physical simulation, etc.) for efficient sampling from complex probability distributions. The core idea is to gradually reduce the noise intensity (simulated annealing process) to guide the sample from a high-noise random distribution to the high-probability region of the target data distribution. S4 is implemented through the following sub-steps: (4.1) Calculate the channel sample correction step size according to the channel posterior score, expressed as follows:

[0056] where is a hyperparameter used to control the step size of the annealed Langevin sampler.

[0057] The channel sample is corrected using the channel sample correction step size and the channel posterior score, and the corrected channel sample is expressed as follows:

[0058]

[0059] where represents the updated channel sample, represents the standard Gaussian white noise.

[0060] (4.2) Calculate the interference sample correction step size according to the interference posterior score, expressed as follows:

[0061] The interference sample is corrected using the interference sample correction step size and the interference posterior score, and the corrected interference sample is expressed as follows:

[0062]

[0063] (4.3) Update the channel prior variance according to the corrected channel sample, expressed as follows:

[0064]

[0065]

[0066] where denotes the updated channel prior variance at time t, denotes the element-wise square.

[0067] S5: Using the method of S3, the posterior score corresponding to the corrected channel sample and interference sample is obtained; the channel sample and interference sample at the next time (i.e. t' time) are obtained by using the inverse process of the diffusion model; t' is updated as t = t + dt, where dt is the inverse time step. S5 is implemented through the following sub-steps: (5.1) According to the corrected channel sample, the channel posterior score corresponding to the corrected channel sample is calculated in the manner of S3, and the expression is as follows:

[0068] In the formula, is the corrected channel sample corresponding channel prior score, is the corrected channel sample corresponding channel conditional score.

[0069] At the same time, according to the corrected interference sample, the interference posterior score corresponding to the corrected interference sample is calculated in the manner of S3, and the expression is as follows:

[0070] In the formula, is the corrected interference sample corresponding interference prior score, is the corrected interference sample corresponding interference conditional score.

[0071] (5.2) The channel sample at the next time is obtained by using the inverse process of the diffusion model, and the expression is as follows:

[0072]

[0073] In the formula, d is the current sampling step, and r is a hyperparameter for controlling the sampling time step, which is set by human.

[0074] At the same time, the interference sample at the next time is obtained by using the inverse process of the diffusion model, and the expression is as follows:

[0075] (5.3) Let t' = t + dt.

[0076] S6: Determine whether t' reaches the set threshold t minIf yes, output the final estimate; if no, update the channel prior variance in the way of step (4.3) according to the channel sample at next time obtained in S5, and return to execute S3-S6.

[0077] The final estimate includes the mean of K channel samples at t' time and the mean of K interference samples The expression is as follows:

[0078]

[0079] Figure 4 The figure shows the curve of the normalized mean square error (NMSE) of the channel estimation using the method of the present application with the change of the signal-to-interference ratio (SIR) when the interference signal is an LFM or HFM signal with unknown parameters. The bandwidth of the communication signal is 4 kHz, and the pilot symbol is a binary phase shift keying (BPSK) symbol. The bandwidth of the interference signal is 1 kHz, and the duration is 1 second. The start frequency and end frequency are random and fall within the frequency range of the single-carrier communication signal. Meanwhile, the phase and arrival time of the signal are also random. Under the same conditions, the method of the present application: diffusion model-based sparse Bayesian learning (DM-SBL) is compared with traditional channel estimation algorithms, including: minimum mean square error estimation (MMSE), orthogonal matching pursuit (OMP), vector approximate message passing (VAMP), sparse Bayesian learning (SBL), and expectation maximization Bernoulli Gaussian generalized approximate message passing (EM-BGGAMP). It can be seen from Figure 4 that compared with the traditional channel estimation algorithms, the method of the present application has achieved significant performance improvement.

[0080] Figure 5The curve of the channel estimation NMSE of the method of the application with the change of SIR when the interference signal is an orthogonal frequency division multiplexing (hereinafter referred to as OFDM) signal with 256 subcarriers is shown. The subcarrier spacing is 15Hz, 64 subcarriers are used for pilots, and the remaining 192 subcarriers are used for data transmission. The pilot symbol is a fixed BPSK symbol, the data symbol is a randomly generated quadrature phase shift keying (hereinafter referred to as QPSK) symbol, and the signal bandwidth falls within the frequency range of the single carrier communication signal. It is known from Figure 5 It can be seen that the method of the application also achieves certain performance improvement compared with the traditional channel estimation algorithm for OFDM interference signals of this kind.

[0081] The embodiment also provides an anti-interference underwater acoustic channel estimation device based on model data double driving, as shown in the figure, the device comprises a memory and one or more processors, the memory stores executable codes, and the one or more processors execute the executable codes to implement the anti-interference underwater acoustic channel estimation method based on model data double driving in the above embodiment. Figure 6

[0082] The anti-interference underwater acoustic channel estimation device based on model data double driving of the application can be applied to any device with data processing capability, which can be a device or apparatus such as a computer. The anti-interference underwater acoustic channel estimation device based on model data double driving can be realized by software, or realized by hardware or a combination of software and hardware. Taking software realization as an example, as a logical device, it is formed by reading the corresponding computer program instructions in the non-volatile memory into the memory and running through the processor of the device with data processing capability where it is located. From the hardware level, as shown in the figure, it is a hardware structure diagram of the device with data processing capability where the anti-interference underwater acoustic channel estimation device based on model data double driving of the application is located. In addition to the processor, the memory, the network interface, and the non-volatile memory shown in the figure, the device with data processing capability where the device of the application is located in the embodiment usually includes other hardware according to the actual functions of the device with data processing capability, and details are not described here. Figure 6 Figure 6

[0083] The implementation process of the functions and roles of each unit in the above device is specifically described in the implementation process of the corresponding steps in the above method, and is not described here.

[0084] ​​​For the device embodiments, since they basically correspond to the method embodiments, the relevant parts are described in the part of the method embodiments. The device embodiments described above are only illustrative, wherein the units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, i.e., can be located in one place or distributed on multiple network units. Some or all of the modules can be selected to achieve the purposes of the present application according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0085] The embodiments of the present application also provide a computer readable storage medium, which stores a program, and the program is executed by a processor to implement the model data double-driven anti-interference underwater acoustic channel estimation method in the above embodiments.

[0086] The computer readable storage medium can be an internal storage unit of any data processing capable device of any of the preceding embodiments, such as a hard disk or a memory. The computer readable storage medium can also be an external storage device of any data processing capable device, such as a plug-in hard disk, a smart media card (SMC), an SD card, a flash card, etc. Further, the computer readable storage medium can include both an internal storage unit and an external storage device of any data processing capable device. The computer readable storage medium is used to store computer programs and other programs and data required by any data processing capable device, and can also be used to temporarily store data that has been output or will be output.

[0087] Those skilled in the art can understand that the above description is only preferred examples of the application, and is not used to limit the application, although the application is described in detail with reference to the foregoing examples, those skilled in the art can still modify the technical solutions recorded in the foregoing examples, or make equivalent replacements for some technical features. Any modification, equivalent replacement, etc. within the spirit and principles of the application shall be included in the protection scope of the application.

Claims

1. A method for anti-interference underwater acoustic channel estimation based on model data double driving, characterized in that, The method comprises the following steps: S1: obtaining an interference sample data set, and training a neural network based on a U-net structure; S2: preprocessing a received underwater acoustic signal to obtain a pilot part signal y after symbol rate sampling, initializing K channel samples h and K interference samples n using Gaussian white noise, initializing a sampling time t'=1, and initializing a channel prior variance γ; S3: setting the sampling time t=t', obtaining an interference prior score at the time t using the neural network, obtaining a channel prior score at the time t using an analytical prior model of the channel samples, obtaining a log-likelihood function at the time t using an approximation method, obtaining an interference conditional score at the time t by taking a gradient of the log-likelihood function with respect to a conjugate of the interference, and obtaining a channel conditional score at the time t by taking a gradient of the log-likelihood function with respect to a conjugate of the channel; obtaining an interference posterior score at the time t by weighted summing the interference prior score and the interference conditional score, and obtaining a channel posterior score at the time t by weighted summing the channel prior score and the channel conditional score; S4: using an annealed Langevin sampling method, correcting the channel samples and the interference samples according to the posterior scores, and updating the channel prior variance according to the corrected channel samples; S5: using the method of S3 to obtain posterior scores corresponding to the corrected channel samples and the interference samples, using an inverse process of a diffusion model to obtain channel samples and interference samples at a next time, and updating t'=t+dt, wherein dt is an inverse time step; S6: judging whether t' reaches a set threshold t min , if yes, outputting a final estimation value; if no, updating the channel prior variance according to the channel sample at the next moment obtained in S5, and returning to execute S3-S6; the final estimation value including the mean value of the K channel samples at t' moment and the mean value of the K interference samples .

2. The model data double-driven anti-interference underwater acoustic channel estimation method according to claim 1, characterized in that, In S1, the neural network comprises an encoder, a decoder, and a skip connection module; real and imaginary parts of input data are stacked to form a two-channel tensor, and are projected to a 64-dimensional feature space through an initial convolution layer to serve as an input of the encoder; The encoder and the decoder each comprise 32 cascaded convolution modules, and each convolution module comprises a convolution layer and a ReLU activation function; The skip connection module stacks output feature maps of each convolution module in the encoder and input feature maps of a corresponding level in the decoder in a channel to enhance gradient flow and retain multi-scale features; The feature maps output by the decoder are projected to a two-dimensional space through a final convolution layer to generate an interference prior score estimation result.

3. The model data double-driven anti-interference underwater acoustic channel estimation method according to claim 1, characterized in that, The real part of the interference prior score The real part of the interference prior score The real part of the interference prior score The real part of the interference prior score channel prior score at time t is obtained by multiplying the channel prior distribution with the channel disturbance kernel function at time t and taking the gradient of the conjugate of the channel at time t.

4. The model data double-driven anti-interference underwater acoustic channel estimation method according to claim 3, characterized in that, The log-likelihood function at the time t is obtained by scaling the interference in a reception model of the pilot part signal with the interference at the time t, adding noise to the scaled interference, scaling the channel with the channel at the time t, and adding noise to the scaled channel.

5. The model data double-driven anti-interference underwater acoustic channel estimation method according to claim 1, characterized in that, S4 is implemented through the following sub-steps: (4.1) calculating a channel sample correction step: the channel correction step is a hyperparameter divided by a square of a two-norm of the channel score; obtaining the modified channel samples is the sum of the product of the original channel samples, the channel modification step size and the channel posterior probability, and the product of twice the square root of the channel modification step size and the standard Gaussian white noise; (4.2) calculating an interference sample correction step: the channel correction step is a hyperparameter divided by a square of a two-norm of the interference score; obtaining the modified interference sample is the sum of the original interference sample, the interference correction step and the product of the square root of twice the interference correction step and the standard Gaussian white noise (4.3) updating the channel prior variance according to the corrected channel samples using an expectation maximization algorithm.

6. The model data double-driven anti-interference underwater acoustic channel estimation method according to claim 1, characterized in that, In the S5, the next time channel sample is obtained by using a diffusion model inverse process, specifically, the next time channel sample is obtained by subtracting a second term from a first term and adding a third term, the first term is the current time corrected channel sample, the second term is a product of a diffusion process drift coefficient and the current time corrected channel sample, half of the product, a sum of a product of the drift coefficient and a channel posterior score, and a product of the sum and an inverse time step, the third term is a product of a square root of a product of a negative drift coefficient and the inverse time step and a standard Gaussian white noise; In the S5, the next time channel sample is obtained by using a diffusion model inverse process, specifically, the next time channel sample is obtained by subtracting a second term from a first term and adding a third term, the first term is the current time corrected channel sample, the second term is a product of a diffusion process drift coefficient and the current time corrected channel sample, half of the product, a sum of a product of the drift coefficient and a channel posterior score, and a product of the sum and an inverse time step, the third term is a product of a square root of a product of a negative drift coefficient and the inverse time step and a standard Gaussian white noise; 7. A model data double-driven anti-interference underwater acoustic channel estimation device, characterized in that, The model data double-driven anti-interference underwater acoustic channel estimation method comprises a memory and one or more processors, the memory stores executable code, and the one or more processors execute the executable code to implement the model data double-driven anti-interference underwater acoustic channel estimation method.

8. A computer-readable storage medium, characterized in that, The model data double-driven anti-interference underwater acoustic channel estimation method comprises a memory and one or more processors, the memory stores executable code, and the one or more processors execute the executable code to implement the model data double-driven anti-interference underwater acoustic channel estimation method.

Images