Channel simulation method and system based on de-noising diffusion probability model

Abstract

The invention provides a channel simulation method and system based on a de-noising diffusion probability model. The method comprises the following steps: training based on a diffusion probability model in combination with real atmospheric turbulence data to obtain a channel simulation model; a random Gaussian noise sample is input into a channel simulation model, so that a Zernike coefficient vector similar to real atmospheric turbulence data distribution can be generated, and a turbulence phase screen closer to a real atmospheric environment is generated according to the generated Zernike coefficient vector, so that the precision and applicability of atmospheric turbulence modeling are improved, and the modeling efficiency is improved. And reliable support is provided for research and application of related technologies.

Description

Technical Field

[0001] The present invention relates to the fields of free space optical communication, optical tracking and adaptive optics technology, and in particular to a channel simulation method and system based on a denoising diffusion probability model. Background Art

[0002] When a light beam propagates through the atmosphere, the phase distribution of the light wave will be significantly distorted due to the refractive index fluctuations caused by atmospheric turbulence, thus affecting the transmission stability and quality of the light beam. This problem is particularly prominent in high-precision fields such as laser communications and astronomical observations. In order to effectively overcome the influence of turbulence, adaptive optical systems are widely used, and accurate modeling of atmospheric turbulence is an important part of designing and optimizing these systems.

[0003] The impact of turbulence on light waves is mainly reflected in phase distortion, so the core task of turbulence modeling is to generate a phase screen that simulates atmospheric turbulence. The current modeling method is generally divided into two steps: first, a power spectrum model of turbulent refractive index is constructed based on theory, and then a phase screen is generated based on the power spectrum. Commonly used power spectrum models include Kolmogorov turbulence spectrum, VonKarman spectrum, and non-Kolmogorov spectrum, while the phase screen generation method usually uses Zernike polynomials or power spectrum inversion technology.

[0004] In the traditional technical solution, taking the Kolmogorov turbulence spectrum as an example, the refractive index power spectrum derived based on the Kolmogorov theory is as follows: ; (1) in, represents the refractive index power spectral density; represents the spatial wave number; Represents the atmospheric refractive index structure constant, which characterizes the turbulence intensity.

[0005] In specific applications, it is necessary to combine field observations or empirical formulas to estimate the value of the atmospheric refractive index structure constant, which is closely related to various factors such as temperature, wind speed, air pressure, altitude, etc. After obtaining the power spectrum and the atmospheric refractive index structure constant, a phase screen is constructed through methods such as Zernike polynomial expansion or power spectrum inversion to simulate the influence of atmospheric turbulence on the phase distribution of the light beam.

[0006] Traditional technical solutions have the following problems: 1. There is a gap between idealized assumptions and the real environment. The classic Kolmogorov turbulence model is based on the assumptions of isotropy and stability of atmospheric turbulence. These assumptions are often difficult to meet in actual scenarios, especially in areas with complex environments or high frequency of changes (such as over cities, mountainous areas, or sponges). These factors will lead to a decrease in simulation accuracy.

[0007] Second, it is difficult to accurately measure and estimate the atmospheric refractive index structure constant. The atmospheric refractive index structure constant is affected by local meteorological conditions (such as temperature, humidity, wind field, etc.) and geographical factors (such as altitude, topography, latitude, etc.). It needs to be measured by special equipment and observation methods, and can also be estimated based on empirical models. However, its accurate acquisition in real time is still a technical difficulty, and the accumulation of errors will also affect the final turbulence simulation effect.

[0008] Therefore, it is difficult for traditional power spectrum models to achieve high-fidelity phase distortion simulation in a changeable real atmospheric environment, which in turn affects the design and performance evaluation of free-space optical communications, optical tracking, and adaptive optical systems.

[0009] In response to the problems of traditional technical solutions, the existing technology proposes an exploratory attempt to combine artificial intelligence with channel simulation based on the Generative Adversarial Network (GAN) method. However, in the early stage of training of the GAN model, the confrontation between the generator network and the discriminator network is unstable, especially under small sample conditions, and there is a high probability of gradient disappearance and gradient explosion problems. In the final evaluation, there is a pattern collapse phenomenon, that is, the model only produces a single sample pattern and has poor generalization. This increases the difficulty and cost of model training and limits the final effect of the model to a certain extent. Summary of the invention

[0010] In view of this, an embodiment of the present invention provides a channel simulation method and system based on a noise reduction diffusion probability model to eliminate or improve one or more defects in the prior art.

[0011] On the one hand, the present invention provides a channel simulation method based on a noise reduction diffusion probability model, the method comprising: Random Gaussian noise samples are input into the pre-trained channel simulation model to generate a Zernike coefficient vector similar to the distribution of real atmospheric turbulence data; The generated Zernike coefficient vector is converted into a phase screen to simulate the influence of atmospheric turbulence on the beam imaging quality; Among them, the channel simulation model adopts a denoised diffusion probability model, including a forward diffusion process and an inverse generation process; the training method of the channel simulation model includes: collecting wavefront distortion data affected by atmospheric turbulence and fitting to obtain a true Zernike coefficient vector to construct a training data set; using the training data set to train the denoised diffusion probability model, in the forward diffusion process, gradually adding Gaussian noise to the true Zernike coefficient vector to generate noise data, in the inverse generation process, using a preset neural network to gradually remove the noise of the noise data to generate a predicted Zernike coefficient vector; constructing a loss function between the noise predicted by the preset neural network and the actual added noise, and optimizing the network parameters by performing gradient descent on the loss function; repeating the forward diffusion process and the inverse generation process until the preset neural network converges to obtain the channel simulation model.

[0012] In some embodiments of the present invention, wavefront distortion data affected by atmospheric turbulence is collected and fitted to obtain a true Zernike coefficient vector to construct a training data set, including: In a real atmospheric turbulence environment, a wavefront detector is used to obtain wavefront distortion data affected by atmospheric turbulence; The wavefront distortion data is orthogonally expanded by Zernike polynomials, and the Zernike coefficient vector of each group of wavefront distortion data is obtained according to the wavefront phase distribution fitting, and the Zernike coefficient vectors are summarized as the training data set.

[0013] In some embodiments of the present invention, in the forward diffusion process, Gaussian noise is gradually added to the true Zernike coefficient vector to generate noise data, including: Gaussian noise is added to the true Zernike coefficient vector at each time step to form a series of Markov chains, calculated as follows: ; in, Indicates The noise data obtained in the first step; represents normal distribution; Indicates The noise variance of the step; represents the identity matrix; The true Zernike coefficient vector is diffused to the noise distribution of any time step by the product of a series of single-step transfer distributions, and the calculation formula is: ; in, Represents the time step The corresponding signal retention coefficient; From time step 1 to The cumulative signal retention coefficient.

[0014] In some embodiments of the present invention, the preset neural network adopts a three-layer U-Net structure.

[0015] In some embodiments of the present invention, in the reverse generation process, a preset neural network is used to gradually remove the noise of the noise data, including: According to the reverse time step, the noise data is updated according to the noise predicted by the preset neural network, and the calculation formula of its Markov transition distribution is: ; in, Indicates The noise data obtained in the first step; represents normal distribution; Represents the preset neural network The mean of the step-by-step forecasts; represents an optional covariance term.

[0016] In some embodiments of the present invention, the calculation formula for the mean of the preset neural network single-step prediction is: ; in, Indicates The noise data obtained in the first step; Represents the time step The corresponding signal retention coefficient; Indicates The noise variance of the step; Indicates the preset neural network prediction at time step Noise added on.

[0017] In some embodiments of the present invention, during the reverse generation process, a preset neural network is used to gradually remove the noise of the noise data to generate a predicted Zernike coefficient vector, and its single-step calculation formula is: ; in, Indicates The noise data obtained in the first step; Represents the time step The corresponding signal retention coefficient; Indicates The noise variance of the step; Indicates the preset neural network prediction at time step Noise added on Represents the time step in the reverse process The corresponding noise standard deviation; represents Gaussian noise; represents normal distribution; Represents the identity matrix.

[0018] In some embodiments of the present invention, constructing a loss function between the noise predicted by the preset neural network and the actually added noise includes: The loss function adopts mean square error loss, and the calculation formula is: ; in, Indicated in Step 1: Given the original data, calculate the mean under the added noise condition; Indicates the preset neural network prediction at time step Noise added on Indicates that at time step The noise added to the real represents the preset neural network parameters.

[0019] In some embodiments of the present invention, converting the generated Zernike coefficient vector into a phase screen includes: The conversion relationship is: ; in, Represents polar coordinates; Indicates Zernike basis functions of order; represents the Zernike polynomial coefficients; Represents the upper limit of the order of the Zernike polynomial.

[0020] On the other hand, the present invention also provides a channel simulation system based on a denoised diffusion probability model, characterized in that when the system is executed, the steps of the channel simulation method based on a denoised diffusion probability model as described in any one of the above are implemented.

[0021] The present invention provides a channel simulation method and system based on a denoised diffusion probability model, including: obtaining a channel simulation model based on a diffusion probability model combined with real atmospheric turbulence data training; inputting random Gaussian noise samples into the channel simulation model, thereby generating a Zernike coefficient vector similar to the distribution of real atmospheric turbulence data, and then generating a turbulence phase screen closer to the real atmospheric environment based on the generated Zernike coefficient vector. The present invention not only overcomes the reliance of existing atmospheric turbulence simulations on idealized assumptions and refractive index structure constant estimation, but also can adaptively capture complex turbulence characteristics under the support of large-scale data, providing more accurate and flexible atmospheric turbulence channel simulations for subsequent adaptive optics, free-space optical communications, optical tracking and other technologies.

[0022] Additional advantages, purposes, and features of the present invention will be described in part in the following description, and will become apparent to those skilled in the art after studying the following, or may be learned from the practice of the present invention. The purposes and other advantages of the present invention may be achieved and obtained by the structures specifically indicated in the specification and the accompanying drawings.

[0023] Those skilled in the art will appreciate that the objectives and advantages that can be achieved with the present invention are not limited to the above specific description, and the above and other objectives that can be achieved by the present invention will be more clearly understood from the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The drawings described herein are used to provide a further understanding of the present invention, constitute a part of the present application, and do not constitute a limitation of the present invention. In the drawings: Figure 1 The figure is a schematic diagram of the steps of a channel simulation method based on a noise reduction diffusion probability model in one embodiment of the present invention.

[0025] Figure 2 Schematic diagram of the working process of the diffusion model in one embodiment of the present invention.

[0026] Figure 3 Schematic diagram of the U-Net neural network structure in one embodiment of the present invention.

[0027] Figure 4 Schematic diagram of a neural network training process in one embodiment of the present invention.

[0028] Figure 5 The figure is a flow chart of a channel simulation method based on a noise reduction diffusion probability model in one embodiment of the present invention. DETAILED DESCRIPTION

[0029] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments and the accompanying drawings. Here, the illustrative embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0030] It should also be noted that, in order to avoid obscuring the present invention due to unnecessary details, only structures and / or processing steps closely related to the solutions according to the present invention are shown in the accompanying drawings, while other details that are not closely related to the present invention are omitted.

[0031] It should be emphasized that the term “include / comprises” when used herein refers to the presence of features, elements, steps or components, but does not exclude the presence or addition of one or more other features, elements, steps or components.

[0032] It should also be noted that, unless otherwise specified, the term “connection” herein may refer not only to a direct connection but also to an indirect connection involving an intermediate.

[0033] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. In the accompanying drawings, the same reference numerals represent the same or similar components, or the same or similar steps.

[0034] It should be emphasized here that the step marks mentioned below are not intended to limit the order of the steps, but it should be understood that the steps can be executed in the order mentioned in the embodiment, or in a different order from the embodiment, or several steps can be executed simultaneously.

[0035] In order to solve the limitations of traditional atmospheric turbulence simulation schemes in idealized assumptions and refractive index structure constant estimation, as well as the problems of high training difficulty, high cost and poor generalization ability of existing simulation schemes based on generative adversarial networks, the present invention provides a channel simulation method based on a denoised diffusion probability model, such as Figure 1 As shown, the method includes the following steps S101-S102: Step S101: inputting random Gaussian noise samples into the pre-trained channel simulation model to generate a Zernike coefficient vector similar to the distribution of real atmospheric turbulence data.

[0036] Step S102: converting the generated Zernike coefficient vector into a phase screen to simulate the influence of atmospheric turbulence on the beam imaging quality.

[0037] In step S101, the channel simulation model adopts a noise-removing diffusion probability model, which includes a forward diffusion process and a reverse generation process.

[0038] The following is a further explanation of the training method of the channel simulation model.

[0039] First, construct a training dataset.

[0040] In a real atmospheric turbulence environment, a wavefront detector is used to collect a large amount of wavefront distortion data affected by atmospheric turbulence. Commonly used wavefront detectors include Hartmann-Shack sensors, shearing interferometers, etc., which can be selected according to different experimental requirements.

[0041] The wavefront distortion data is orthogonally expanded by Zernike polynomials, which are a set of polynomial bases that are complete and mutually orthogonal within the unit circle, as shown in formula (2): ; (2) in, Represents polar coordinates; Indicates Zernike basis functions of order; represents the Zernike polynomial coefficients; Represents the upper limit of the order of the Zernike polynomial.

[0042] The energy of the Zernike coefficients is mainly concentrated in the low- and medium-order modes, and the energy of the high-order modes is very small. Based on existing research, for small and medium-aperture systems, optical correction mainly focuses on low-order and some medium-order aberrations, and the reference range of the Zernike mode number can be M=28~66; for medium and large-aperture adaptive optical systems, the reference range of M is 66~231; as for large-aperture high-end adaptive optical systems, the corresponding mode number M can reach 496 or more. The specific value can be adjusted according to actual needs. In one embodiment of the present invention, M=64, which is more suitable for general medium-sized optical systems, is used, which is sufficient to include most of the low- and medium-order aberrations.

[0043] According to the measured wavefront phase distribution, the Zernike coefficient vector of each set of wavefront distortion data is fitted with the help of Gram-Schmidt orthogonal algorithm, least squares method and other common fitting methods, and the fitted Zernike coefficient vectors are summarized as the training data set.

[0044] Secondly, the model is constructed. The present invention uses a denoised diffusion probability model to generate Zernike coefficients that conform to the real atmospheric turbulence characteristics. Figure 2 As shown in Figure 3, the model consists of two parts: forward diffusion process (Forward Diffusion) and reverse generation process (Reverse Generation).

[0045] In the forward diffusion process, Gaussian noise is gradually added to the Zernike coefficient vector obtained by the above fitting according to the predefined noise parameters, forming a series of Markov chains from clear data to completely noisy data.

[0046] In some embodiments, Gaussian noise is added to the true Zernike coefficient vector at each time step to form a series of Markov chains. The Markov process can be positioned as formula (3): ; (3) in, Indicates The noise data obtained in the first step; represents normal distribution; Indicates The noise variance of the step; Represents the identity matrix.

[0047] By multiplying a series of single-step transfer distributions, we can obtain the noise distribution of the true Zernike coefficient vector diffused to any time step. The calculation formula is shown in formula (4): ; (4) in, Represents the time step The corresponding signal retention coefficient; From time step 1 to The cumulative signal retention coefficient is calculated as shown in formulas (5) and (6): ; (5) ; (6) In practical applications, linear or nonlinear settings can be used as needed. , which can also be set as the parameter to be learned. The commonly used reference range of magnitude is Preferably, in one embodiment of the present invention, Set to equally spaced linear interpolation between 0.001 and 0.05.

[0048] In this way, the noise data of any time step can be obtained. That is, by randomly selecting a time step and injecting noise into the true Zernike coefficient vector, training samples with different degrees of noise can be obtained at different time steps, which are used to train the noise prediction ability of the preset neural network.

[0049] In the reverse generation process, the noise added at each time step is predicted by a preset neural network to gradually remove the noise added in the forward diffusion process and restore the predicted Zernike coefficient vector that is similar to the distribution of the true Zernike coefficient vector.

[0050] In some embodiments, Figure 3 As shown, the preset neural network adopts a three-layer U-Net structure for noise prediction.

[0051] The inverse generation process starts from a pure Gaussian noise sample Starting from Iterate backward to 1. At each step, the data of the next time step is predicted based on the noise predicted by the preset neural network.

[0052] In some embodiments, the above Markov transition distribution is defined as formula (7): ; (7) in, Indicates The noise data obtained in the first step; represents normal distribution; Represents the preset neural network The mean of the step-by-step forecasts; represents an optional covariance term. In this embodiment, for simplicity, Fixed to a given function, such as ,in, Desirable Or the interpolation result as shown in formula (8).

[0053] ; (8) In some embodiments, when the time step When , it is necessary to add some Gaussian noise to the predicted mean. By continuously iterating this inverse process, the predicted Zernike coefficient vector is finally obtained, which is an output with a distribution similar to the true Zernike coefficient vector.

[0054] Based on the definition of the forward diffusion process, the calculation formula of the preset neural network single-step reverse mean term is shown in formula (9): ; (9) in, Indicates The noise data obtained in the first step; Represents the time step The corresponding signal retention coefficient; Indicates The noise variance of the step; Represents the preset neural network prediction at time step Noise added on.

[0055] Using formula (9), the model can remove the noise in the noisy data and gradually approach the real data distribution.

[0056] Based on the above model structure and working principle, finally, a large number of training data sets are used to train the model. The overall training process is as follows: Figure 4 shown.

[0057] During the training process, a loss function is constructed between the noise predicted by the preset neural network and the actual added noise, and the network parameters are optimized by gradient descent of the loss function. The forward diffusion process and the reverse generation process are repeated until the preset neural network converges to obtain the channel simulation model. The trained channel simulation model can reversely denoise from random Gaussian noise and generate Zernike coefficients with a similar distribution to the training data set.

[0058] In some embodiments, the loss function uses mean square error loss, and the calculation formula is shown in formula (10): ; (10) in, Indicated in Step 1: Given the original data, calculate the mean under the added noise condition; Represents the preset neural network prediction at time step Noise added on Indicates that at time step The noise added to the real Represents the preset neural network parameters.

[0059] like Figure 5 As shown, it is a flowchart of realizing sampling generation using the trained channel simulation model.

[0060] In step S101, from At the beginning, the noise is gradually removed by using the reverse diffusion process of the neural network until the Zernike coefficient vector is generated. The single-step calculation is shown in formula (11): ; (11) in, Indicates The noise data obtained in the first step; Represents the time step The corresponding signal retention coefficient; Indicates The noise variance of the step; Represents the preset neural network prediction at time step Noise added on Represents the time step in the reverse process The corresponding noise standard deviation; represents Gaussian noise; represents normal distribution; Represents the identity matrix.

[0061] The final predicted Zernike coefficient vector can be regarded as a reconstruction result with a distribution close to that of the true Zernike coefficient vector (real atmospheric turbulence data), and its dimension is consistent with that of the true Zernike coefficient vector.

[0062] In step S102, since the Zernike polynomial can approximate any reasonable phase distribution within the unit circle, after obtaining the Zernike coefficients of the required order, it is substituted into the above formula (2) and the phase screen can be obtained by using the conversion relationship: , which can be further used to simulate the real effect of atmospheric turbulence on the light beam.

[0063] Corresponding to the above method, the present invention also provides a channel simulation system based on a noise-removing diffusion probability model, which, when executed, implements the steps of any of the channel simulation methods based on a noise-removing diffusion probability model mentioned above.

[0064] Corresponding to the above method, the present invention also provides an electronic device, which includes a computer device, the computer device includes a processor and a memory, the memory stores computer instructions, the processor is used to execute the computer instructions stored in the memory, and when the computer instructions are executed by the processor, the electronic device implements the steps of the method described above.

[0065] The embodiment of the present invention also provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the steps of the aforementioned edge computing server deployment method are implemented. The computer-readable storage medium can be a tangible storage medium, such as a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a floppy disk, a hard disk, a removable storage disk, a CD-ROM, or any other form of storage medium known in the technical field.

[0066] It should be understood by those skilled in the art that the exemplary components, systems and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software or a combination of the two. Whether it is performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention. When implemented in hardware, it can be, for example, an electronic circuit, an application-specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, etc. When implemented in software, the elements of the present invention are programs or code segments used to perform the required tasks. The program or code segment can be stored in a machine-readable medium, or transmitted on a transmission medium or a communication link via a data signal carried in a carrier.

[0067] It should be clear that the present invention is not limited to the specific configuration and processing described above and shown in the figures. For the sake of simplicity, a detailed description of the known method is omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between the steps after understanding the spirit of the present invention.

[0068] In the present invention, features described and / or illustrated for one embodiment may be used in the same or similar manner in one or more other embodiments, and / or combined with features of other embodiments or replace features of other embodiments.

[0069] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the embodiments of the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A channel simulation method based on a noise-removing diffusion probability model, characterized in that: The method comprises: Random Gaussian noise samples are input into the pre-trained channel simulation model to generate a Zernike coefficient vector similar to the distribution of real atmospheric turbulence data; The generated Zernike coefficient vector is converted into a phase screen to simulate the influence of atmospheric turbulence on the beam imaging quality; Among them, the channel simulation model adopts a denoised diffusion probability model, including a forward diffusion process and an inverse generation process; the training method of the channel simulation model includes: collecting wavefront distortion data affected by atmospheric turbulence and fitting to obtain a true Zernike coefficient vector to construct a training data set; using the training data set to train the denoised diffusion probability model, in the forward diffusion process, gradually adding Gaussian noise to the true Zernike coefficient vector to generate noise data, in the inverse generation process, using a preset neural network to gradually remove the noise of the noise data to generate a predicted Zernike coefficient vector; constructing a loss function between the noise predicted by the preset neural network and the actual added noise, and optimizing the network parameters by performing gradient descent on the loss function; repeating the forward diffusion process and the inverse generation process until the preset neural network converges to obtain the channel simulation model.

2. The channel simulation method based on the noise reduction diffusion probability model according to claim 1 is characterized in that: The wavefront distortion data affected by atmospheric turbulence is collected and fitted to obtain the true Zernike coefficient vector to construct a training data set, including: In a real atmospheric turbulence environment, a wavefront detector is used to obtain wavefront distortion data affected by atmospheric turbulence; The wavefront distortion data is orthogonally expanded by Zernike polynomials, and the Zernike coefficient vector of each group of wavefront distortion data is obtained according to the wavefront phase distribution fitting, and the Zernike coefficient vectors are summarized as the training data set.

3. The channel simulation method based on the noise reduction diffusion probability model according to claim 1 is characterized in that: In the forward diffusion process, Gaussian noise is gradually added to the true Zernike coefficient vector to generate noise data, including: Gaussian noise is added to the true Zernike coefficient vector at each time step to form a series of Markov chains, calculated as follows: ； in, Indicates The noise data obtained in the first step; represents normal distribution; Indicates The noise variance of the step; represents the identity matrix; The true Zernike coefficient vector is diffused to the noise distribution of any time step by the product of a series of single-step transfer distributions, and the calculation formula is: ； in, Represents the time step The corresponding signal retention coefficient; From time step 1 to The cumulative signal retention factor.

4. The channel simulation method based on the noise reduction diffusion probability model according to claim 1, characterized in that: The preset neural network adopts a three-layer U-Net structure.

5. The channel simulation method based on the noise reduction diffusion probability model according to claim 1, characterized in that: In the reverse generation process, a preset neural network is used to gradually remove the noise of the noise data, including: According to the reverse time step, the noise data is updated according to the noise predicted by the preset neural network, and the calculation formula of its Markov transition distribution is: ； in, Indicates The noise data obtained in the first step; represents normal distribution; Represents the preset neural network The mean of the step-by-step forecasts; represents an optional covariance term.

6. The channel simulation method based on the noise reduction diffusion probability model according to claim 5 is characterized in that: The calculation formula of the mean value of the preset neural network single-step prediction is: ； in, Indicates The noise data obtained in the first step; Represents the time step The corresponding signal retention coefficient; Indicates The noise variance of the step; Indicates the preset neural network prediction at time step Noise added on.

7. The channel simulation method based on the noise reduction diffusion probability model according to claim 1, characterized in that: In the reverse generation process, a preset neural network is used to gradually remove the noise of the noise data to generate a predicted Zernike coefficient vector, and its single-step calculation formula is: ； in, Indicates The noise data obtained in the first step; Represents the time step The corresponding signal retention coefficient; Indicates The noise variance of the step; Indicates the preset neural network prediction at time step Noise added on Represents the time step in the reverse process The corresponding noise standard deviation; represents Gaussian noise; represents normal distribution; Represents the identity matrix.

8. The channel simulation method based on the noise reduction diffusion probability model according to claim 1 is characterized in that: Constructing a loss function between the noise predicted by the preset neural network and the actually added noise, including: The loss function adopts mean square error loss, and the calculation formula is: ； in, Indicated in Step 1: Given the original data, calculate the mean under the added noise condition; Indicates the preset neural network prediction at time step Noise added on Indicates that at time step The noise added to the real represents the preset neural network parameters.

9. The channel simulation method based on the noise reduction diffusion probability model according to claim 1, characterized in that: The generated Zernike coefficient vector is converted into a phase screen, including: The conversion relationship is: ； in, Represents polar coordinates; Indicates Zernike basis functions of order; represents the Zernike polynomial coefficients; Represents the upper limit of the order of the Zernike polynomial.

10. A channel simulation system based on a noise-removing diffusion probability model, characterized in that: When the system is executed, the steps of the channel simulation method based on the noise reduction diffusion probability model as claimed in any one of claims 1 to 9 are implemented.

Images