A method, medium and device for MIMO scattering channel estimation

Abstract

The application discloses a MIMO scattering channel estimation method, medium and equipment, and belongs to the technical field of channel estimation. The method comprises the following steps: obtaining a noisy scattering channel matrix, performing reverse diffusion denoising processing on the basis of a trained DT denoising model according to the noisy scattering channel matrix, and outputting a denoised scattering channel matrix; and using a training data set, a verification data set and a test data set to train the DT denoising model to output the denoised scattering channel matrix through an iterative method and a reverse denoising process. The application solves the defects that the prior art has not effectively applied the diffusion model to the tropospheric scattering channel estimation field and has not proposed a special generative model architecture optimized for the scene characteristics.

Description

Technical Field

[0001] This invention relates to a MIMO scattering channel estimation method, medium, and device, belonging to the field of channel estimation technology. Background Technology

[0002] Tropospheric scattering communication is an important communication method in military and emergency communications, characterized by high reliability, long single-hop distance, strong confidentiality, and strong resistance to environmental interference. To meet the demands of future digital battlefields for greater communication capacity, massively multi-input multiple-output (MIMO) technology has been introduced into tropospheric scattering communication to enhance its capacity. However, the significant spatial loss and severe multipath effects in tropospheric scattering communication links degrade communication quality, necessitating more accurate channel estimation algorithms to compensate for signal fading and improve communication quality. Currently, tropospheric scattering channel estimation algorithms are somewhat lacking in accuracy and computational complexity, highlighting the urgent need for novel and effective channel estimation algorithms to improve performance.

[0003] With the application of Artificial Intelligence (AI) in communication systems, AI-based channel estimation algorithms have demonstrated advantages such as high accuracy and low computational complexity. Generative models, as a representative technology of the new era of AI, have shown great potential in image processing and data recovery. Typical generative models include Generative Adversarial Networks (GANs), Variational Autoencoders (VEAs), and Diffusion Models (DM). Unlike other generative models, DM excels in high-dimensional data generation tasks due to its progressive denoising mechanism, and its underlying mechanism is highly similar to the goal of communication system channels: denoising and reconstructing information signals from noisy signals. The DM generation process typically employs the UNet architecture, but UNet's local feature modeling capability is relatively weak, limiting the model's performance in extracting spatial information from the data.

[0004] In summary, existing technologies have not effectively applied diffusion models (DM) to the field of tropospheric scattering channel estimation, nor have they proposed a dedicated generative model architecture optimized for the characteristics of this scenario. Summary of the Invention

[0005] The purpose of this invention is to provide a MIMO scattering channel estimation method, medium, and device. The DT denoising model is trained through an iterative method and a reverse denoising process to output a denoised scattering channel matrix. This addresses the shortcomings of existing technologies that have not effectively applied the diffusion model (DM) to the field of tropospheric scattering channel estimation, nor have they proposed a dedicated generative model architecture optimized for the characteristics of this scenario.

[0006] To solve the above-mentioned technical problems, the present invention is implemented using the following technical solution:

[0007] In a first aspect, the present invention provides a MIMO scattering channel estimation method, comprising:

[0008] Obtain the noisy scattering channel matrix;

[0009] Based on the noisy scattering channel matrix, back diffusion denoising is performed using the trained DT denoising model to output the denoised scattering channel matrix.

[0010] The training method for the DT denoising model includes:

[0011] The scattering channel matrix is ​​forward diffused using a tropospheric scattering channel model, and standard Gaussian noise is gradually added until the scattering channel matrix evolves into pure noise. Training and validation datasets are established by recording the channel state matrix corresponding to the current time step.

[0012] The received signal is obtained by adding Gaussian white noise to the transmitted pilot signal. The noisy scattering channel matrix is ​​obtained by using the least squares method based on the received signal and the transmitted pilot signal, and a test dataset is established.

[0013] Using the training dataset, validation dataset, and test dataset, the DT denoising model is trained through an iterative method and a reverse denoising process to output the denoised scattering channel matrix.

[0014] Furthermore, the scattering channel matrix is ​​forward diffused using a tropospheric scattering channel model, and standard Gaussian noise is gradually added until the scattering channel matrix evolves into pure noise. Training and validation datasets are established by recording the channel state matrix corresponding to the current time step, including:

[0015] use The tapped channel model architecture is constructed by coupling a tapped delay line structure with an independent Rayleigh fading generator to build a spatially uncorrelated tropospheric scattering channel model.

[0016] Based on the spatially uncorrelated tropospheric scattering channel model, an exponential correlation model is introduced to simulate the correlation between transmit and receive antenna array elements to construct a spatially correlated tropospheric scattering channel model.

[0017] Using the forward diffusion process of the DT denoising model, standard Gaussian noise is gradually added to the scattering channel data in the tropospheric scattering channel model with spatial correlation, and the channel state matrix corresponding to the current time step is recorded to obtain the noisy channel data group.

[0018] Based on the noisy channel data set, training and validation datasets are established by dividing the channel state matrix according to time steps.

[0019] Furthermore, the spatially uncorrelated tropospheric scattering channel model is expressed as:

[0020] ;

[0021] In the formula, This represents a spatially uncorrelated tropospheric scattering channel model. , Indicates all dimensions as The space of complex matrices, This indicates the number of transmit antennas at the transmitter T. Indicates the receiving end The number of receiving antennas, Indicates the first Channel matrix of scattering paths, Indicates the first Power of each scattering path Napier's constant is represented by , and L represents the number of scattering paths. Indicates the carrier frequency. Indicates the first The time delay of each scattering path;

[0022] The spatially correlated tropospheric scattering channel model is expressed as follows:

[0023] ;

[0024] In the formula, This represents a spatially correlated tropospheric scattering channel model. and They represent the sending end. The correlation coefficient matrix between the receiver and the receiver;

[0025] The formula for generating the channel state matrix corresponding to the current time step is expressed as:

[0026] ;

[0027] In the formula, Indicates the current time step The corresponding channel state matrix, Indicates the current time step Cumulative noise scaling factor Indicates the current time step The corresponding channel state matrix at that time, Indicates standard Gaussian noise. Indicates the current time step The noise scaling factor, Indicates the current time step noise factor Indicates from arrive The product of two digits.

[0028] Furthermore, the received signal is obtained by adding Gaussian white noise to the transmitted pilot signal. Based on the received signal and the transmitted pilot signal, the noisy scattering channel matrix is ​​obtained using the least squares method. A test dataset is then established, including:

[0029] Gaussian white noise is added to the transmitted pilot signal to obtain the received signal;

[0030] The scattering channel matrix is ​​decoupled from the transmitted pilot signal using the least squares method to obtain the noisy scattering channel matrix;

[0031] A test dataset was constructed using the noisy scattering channel matrix as the source data and the channel state matrix as the label.

[0032] Furthermore, the received signal is represented as:

[0033] ;

[0034] In the formula, Indicates receiving signal, This represents a spatially correlated tropospheric scattering channel model. Indicates the transmission of pilot signals. And satisfy A denotes the conjugate transpose of the matrix. Indicates all dimensions as The space of complex matrices, This indicates the number of transmit antennas at the transmitter T. Indicates the length of the transmitted pilot. Let N denote the identity matrix, and let N denote additive white Gaussian noise.

[0035] The source data is represented as follows:

[0036] ;

[0037] In the formula, Indicates source data, This represents the conjugate transpose of the pilot signal.

[0038] Furthermore, the DT denoising model is constructed based on the Transformer architecture. Before performing back-diffusion denoising processing based on the trained DT denoising model according to the noisy scattering channel matrix, the following steps are also included:

[0039] The noisy scattering channel matrix is ​​divided into two channels, real and imaginary, and then transformed into a high-dimensional feature vector through linear embedding.

[0040] The time step corresponding to the noisy scattering channel matrix is ​​converted into a set of time position codes using the cofactor selection function;

[0041] The high-dimensional feature vector is added to the temporal location encoding to form the feature enhancement vector, which is then input into the trained DT denoising model.

[0042] Furthermore, the DT denoising model is trained using an iterative method and a reverse denoising process to output the denoised scattering channel matrix, including:

[0043] Determine the starting time step of the reverse denoising process;

[0044] The diffusion signal-to-noise ratio at the current time step is calculated based on the noise scaling factor accumulated at the current time step.

[0045] Based on the diffusion signal-to-noise ratio of the initial time step and the diffusion signal-to-noise ratio of the current time step, determine the time step that is closest to the diffusion signal-to-noise ratio;

[0046] The feature enhancement vector of the noisy scattering channel matrix after adding channel state labels is used as the initial value for the reverse denoising process and input into the DT denoising model.

[0047] Using the DT denoising model, starting from the closest time step, noise estimation is performed step by step on the noisy scattering channel matrix based on the feature enhancement vector to obtain the predicted noise.

[0048] A loss function is defined with the normalized mean square error between the real noise and the predicted noise as the optimization objective. Based on the loss function, the back-diffusion denoising process is used to iteratively remove noise from the noisy scattering channel matrix, and the denoised scattering channel matrix is ​​output.

[0049] Furthermore, the loss function is expressed as:

[0050] ;

[0051] in, Represents the loss function. For the network parameters of the Transformer architecture, Indicates the current time step Current time step Channel state matrix and current time step Real noise Expected value Indicates the current time step Channel state matrix and current time step The noise estimation function, This represents the square of the L2 norm.

[0052] In a second aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that, when the computer program is executed by a processor, it implements the MIMO scattering channel estimation method as described in the first aspect.

[0053] Thirdly, the present invention provides a computer device, characterized in that it comprises:

[0054] Memory, used to store instructions;

[0055] A processor is configured to execute the instructions, causing the device to perform operations implementing the MIMO scattering channel estimation method as described in the first aspect.

[0056] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0057] 1. This invention employs a DT denoising model to perform back-diffusion denoising on noisy scattering channel matrices, effectively suppressing noise interference. By learning the noise distribution characteristics, the DT denoising model gradually recovers the original features of the channel matrix during the back-diffusion denoising process. Compared to traditional estimation methods, this significantly reduces the normalized mean square error of channel estimation and improves estimation accuracy. Furthermore, this invention simulates the forward diffusion process of the channel matrix using a tropospheric scattering channel model, gradually adding Gaussian noise until the channel matrix completely evolves into pure noise. This allows the DT denoising model to learn the complete denoising path from pure noise to a clean channel matrix, thus exhibiting stronger noise resistance in practical applications, especially suitable for low signal-to-noise ratio (SNR) scenarios.

[0058] 2. This invention also utilizes a tropospheric scattering channel model to generate training and validation datasets, which can fully simulate complex channel characteristics such as multipath scattering and time-varying fading in actual communication. By recording the channel state matrix at different time steps, the DT denoising model can learn the time-varying evolution law of the channel matrix, thereby better adapting to the dynamically changing scattering channel environment. By adding Gaussian white noise to the actual pilot signal to generate the test dataset, it more closely resembles the real communication scenario. Combined with the inverse denoising capability of the DT denoising model, this invention can be directly applied to broadband wireless communication systems such as 5G / 6G, improving the performance of MIMO technology in actual deployment and providing technical support for high-reliability, low-latency communication.

[0059] 3. This invention constructs a spatially correlated tropospheric scattering channel model and accurately simulates the multipath scattering characteristics and inter-antenna element correlation of the actual channel based on tapped delay lines and exponential correlation models. The training dataset can fully cover the diffusion process of the channel matrix from a clean state to pure noise. Combined with the DT denoising model of the Transformer architecture, the feature expression capability is enhanced by real-imaginary part separation processing and time position encoding, so that the reverse denoising process can accurately learn the mapping relationship between noise distribution and channel features.

[0060] 4. This invention achieves automatic learning of the joint distribution of noise scaling factor and channel state by iteratively optimizing the loss function, thus overcoming the dependence of traditional methods on explicit noise statistical characteristics. In the test dataset, even in the face of pilot contamination or channel correlation mismatch scenarios, it can still quickly restore channel capacity, significantly improving the deployment robustness of 5G / 6G massive MIMO systems in complex scattering environments. Attached Figure Description

[0061] Figure 1 This is a flowchart illustrating a MIMO scattering channel estimation method provided in an embodiment of the present invention;

[0062] Figure 2 This is a schematic diagram of the forward diffusion of the DT denoising model provided in the embodiment of the present invention.

[0063] Figure 3 This is a schematic diagram of the reverse diffusion denoising process of the DT denoising model provided in the embodiment of the present invention.

[0064] Figure 4 This is a schematic diagram comparing the normalized mean square error results of the DT denoising model and the least squares estimation method provided in this embodiment of the invention. Detailed Implementation

[0065] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0066] Example 1

[0067] like Figure 1 As shown in the figure, this embodiment introduces a MIMO scattering channel estimation method, including:

[0068] Step 1: Obtain the noisy scattering channel matrix.

[0069] The noisy scattering channel matrix obtained in this invention is the input data of the DT denoising model. Its noise characteristics must be consistent with the training dataset to ensure that the DT denoising model can effectively remove noise.

[0070] Step 2: Based on the noisy scattering channel matrix, perform back-diffusion denoising based on the trained DT denoising model, and output the denoised scattering channel matrix.

[0071] The DT denoising model, based on a diffusion model framework, learns the mapping relationship from a noisy channel matrix to a clean channel matrix by simulating the gradual addition and removal of noise. During the training phase, this invention simulates the evolution of the channel matrix from noise-free to completely noisy by progressively adding Gaussian noise to the clean channel matrix. In the denoising phase, the DT denoising model starts from the completely noisy channel matrix and gradually removes noise to recover a clean channel matrix. The denoising process relies on the noise distribution and channel characteristics learned by the DT denoising model during training. Through a reverse denoising process, this invention enables the DT denoising model to accurately estimate and remove noise components from the channel matrix, outputting estimation results close to the real channel.

[0072] In this invention, the training method of the DT denoising model includes:

[0073] The scattering channel matrix is ​​forward diffused using a tropospheric scattering channel model, and standard Gaussian noise is gradually added until the scattering channel matrix evolves into pure noise. Training and validation datasets are established by recording the channel state matrix corresponding to the current time step.

[0074] This invention utilizes a tropospheric scattering channel model to simulate multipath scattering and inter-antenna element correlation in wireless channels, generating a channel matrix that closely approximates the real channel. Furthermore, by progressively adding Gaussian noise to the channel matrix, it simulates the state of the channel matrix under different noise levels, generating a dataset for training and validation.

[0075] The received signal is obtained by adding Gaussian white noise to the transmitted pilot signal. The noisy scattering channel matrix is ​​obtained by using the least squares method based on the received signal and the transmitted pilot signal, and a test dataset is established.

[0076] This invention adds Gaussian white noise to the known pilot signal at the transmitting end to simulate noise interference in actual communication. By utilizing the difference between the received signal and the transmitted pilot signal, the noisy channel matrix is ​​estimated, and channel state labels are added to evaluate model performance. The introduction of Gaussian white noise is to simulate unpredictable noise interference in real wireless communication environments, making the test data closer to real-world application scenarios. By actively adding noise, this invention enables the test dataset to reflect the noise characteristics of real channels, improving the model's generalization ability.

[0077] Using the training dataset, validation dataset, and test dataset, the DT denoising model is trained through an iterative method and a reverse denoising process to output the denoised scattering channel matrix.

[0078] This invention continuously optimizes the network parameters of the DT denoising model by comparing the difference between the predicted noise and the actual noise, enabling the DT denoising model to remove noise more accurately. Specifically, this invention selects the most suitable denoising starting point based on the noise level of the channel matrix, ensuring that the DT denoising model starts denoising from the position closest to the actual channel state. This allows the Transformer-based DT model to capture the time-varying characteristics of the channel matrix and focuses on key channel features through an attention mechanism, thereby improving denoising performance.

[0079] Example 2

[0080] Based on the same inventive concept as Embodiment 1, this embodiment introduces a MIMO scattering channel estimation method, including:

[0081] This embodiment introduces a MIMO scattering channel estimation method, including:

[0082] Step 1: Obtain the noisy scattering channel matrix.

[0083] This embodiment constructs the DT denoising model based on the Transformer architecture. Before performing back-diffusion denoising based on the trained DT denoising model according to the noisy scattering channel matrix, the method further includes:

[0084] Step 1.1: Divide the noisy scattering channel matrix into two channels, real and imaginary, and convert them into high-dimensional feature vectors through linear embedding.

[0085] Step 1.2: Use the cofactor selection function to convert the time step corresponding to the noisy scattering channel matrix into a set of time position codes.

[0086] Step 1.3: Add the high-dimensional feature vector to the temporal location encoding to obtain the feature enhancement vector, and input it into the trained DT denoising model.

[0087] Step 2: Based on the noisy scattering channel matrix, perform back-diffusion denoising processing based on the trained DT denoising model, and output the denoised scattering channel matrix.

[0088] In this embodiment, the training method of the DT denoising model includes:

[0089] Step 2.1: The scattering channel matrix is ​​forward diffused using a tropospheric scattering channel model, with standard Gaussian noise gradually added until the scattering channel matrix evolves into pure noise. Training and validation datasets are established by recording the channel state matrix corresponding to the current time step, including:

[0090] Step 2.1.1: Use The tapped channel model architecture, coupled with an independent Rayleigh fading generator through a tapped delay line structure, constructs a spatially uncorrelated tropospheric scattering channel model.

[0091] In this embodiment, the spatially uncorrelated tropospheric scattering channel model is represented as follows:

[0092] ;

[0093] In the formula, This represents a spatially uncorrelated tropospheric scattering channel model. , Indicates all dimensions as The space of complex matrices, This indicates the number of transmit antennas at the transmitter T. Indicates the receiving end The number of receiving antennas, Indicates the first Channel matrix of scattering paths, Indicates the first Power of each scattering path Napier's constant is represented by , and L represents the number of scattering paths. Indicates the carrier frequency. Indicates the first The time delay of each scattering path.

[0094] Step 2.1.2: Based on the spatially uncorrelated tropospheric scattering channel model, an exponential correlation model is introduced to simulate the correlation between transmit and receive antenna array elements and construct a spatially correlated tropospheric scattering channel model.

[0095] In this embodiment, the spatially correlated tropospheric scattering channel model is represented as follows:

[0096] ;

[0097] In the formula, This represents a spatially correlated tropospheric scattering channel model. and They represent the sending end. The correlation coefficient matrix between the receiver and the receiver.

[0098] Step 2.1.3: Using the forward diffusion process of the DT denoising model, standard Gaussian noise is gradually added to the scattering channel data in the spatially correlated tropospheric scattering channel model, and the channel state matrix corresponding to the current time step is recorded to obtain the noisy channel data set. The forward diffusion process of the DT denoising model is as follows: Figure 2 As shown.

[0099] In this embodiment, the formula for generating the channel state matrix corresponding to the current time step is expressed as:

[0100] ;

[0101] In the formula, Indicates the current time step The corresponding channel state matrix, Indicates the current time step Cumulative noise scaling factor Indicates the current time step The corresponding channel state matrix at that time, Indicates standard Gaussian noise. Indicates the current time step The noise scaling factor, Indicates the current time step noise factor Indicates from arrive The product of two digits.

[0102] Step 2.1.4: Based on the noisy channel data set, establish the training dataset and validation dataset by dividing the channel state matrix according to the time step.

[0103] Step 2.2: Obtain the received signal by adding Gaussian white noise to the transmitted pilot signal. Based on the received signal and the transmitted pilot signal, use the least squares method to obtain the noisy scattering channel matrix and establish a test dataset, including:

[0104] Step 2.2.1: Add Gaussian white noise to the transmitted pilot signal to obtain the received signal.

[0105] In this embodiment, the received signal is represented as:

[0106] ;

[0107] In the formula, Indicates receiving signal, This represents a spatially correlated tropospheric scattering channel model. Indicates the transmission of pilot signals. And satisfy A denotes the conjugate transpose of the matrix. Indicates all dimensions as The space of complex matrices, Indicates the sending end The number of transmitting antennas, Indicates the length of the transmitted pilot. Let represent the identity matrix, and N represent additive white Gaussian noise.

[0108] Step 2.2.2: Decouple the scattering channel matrix from the transmitted pilot signal using the least squares method to obtain the noisy scattering channel matrix.

[0109] Step 2.2.3: Construct a test dataset using the noisy scattering channel matrix as the source data and the channel state matrix as the label.

[0110] In this embodiment, the source data is represented as:

[0111] ;

[0112] In the formula, Indicates source data, This represents the conjugate transpose of the pilot signal.

[0113] Step 2.3: Using the training dataset, validation dataset, and test dataset, train the DT denoising model through an iterative method and a reverse denoising process to output the denoised scattering channel matrix.

[0114] In this embodiment, the DT denoising model is trained using an iterative method and a reverse denoising process to output the denoised scattering channel matrix, including:

[0115] Step 2.3.1: Determine the starting time step of the reverse denoising process.

[0116] Step 2.3.2: Calculate the diffusion signal-to-noise ratio of the current time step based on the noise scaling factor accumulated at the current time step.

[0117] Step 2.3.3: Determine the time step that is closest to the diffusion signal-to-noise ratio based on the diffusion signal-to-noise ratio of the initial time step and the diffusion signal-to-noise ratio of the current time step.

[0118] Step 2.3.4: Use the feature enhancement vector of the noisy scattering channel matrix after adding channel state labels as the initial value for the reverse denoising process and input it into the DT denoising model.

[0119] Step 2.3.5: Using the DT denoising model, starting from the closest time step, the noise of the noisy scattering channel matrix is ​​estimated step by step according to the feature enhancement vector to obtain the predicted noise.

[0120] Step 2.3.6: Define a loss function with the normalized mean square error between the real noise and the predicted noise as the optimization objective. Based on the loss function, iteratively remove noise from the noisy scattering channel matrix using the backdiffusion denoising process, and output the denoised scattering channel matrix. The backdiffusion denoising process of the DT denoising model is as follows: Figure 3 As shown.

[0121] In this embodiment, the loss function is expressed as:

[0122] ;

[0123] in, Represents the loss function. For the network parameters of the Transformer architecture, Indicates the current time step Current time step Channel state matrix and current time step Real noise Expected value Indicates the current time step Channel state matrix and current time step The noise estimation function, This represents the square of the L2 norm.

[0124] The diagram showing the comparison of the normalized mean square error results between the DT denoising model and the least squares estimation method provided in this embodiment of the invention is shown below. Figure 4 As shown, by Figure 4 It can be seen that the normalized mean square error of the predicted noise and the actual noise from 0dB to 30dB is compared, and it can be seen that the DT model of the present invention is better than the least squares estimation in terms of tropospheric scattering channel estimation.

[0125] Example 3

[0126] Based on the same inventive concept as other embodiments, this embodiment describes a computer-readable storage medium having computer instructions stored thereon, which, when executed by a processor, implement the steps of the methods of Embodiment 1 or 2 described above.

[0127] Example 4

[0128] Based on the same inventive concept as other embodiments, this embodiment describes a computer device, including:

[0129] Memory, used to store instructions;

[0130] A processor is configured to execute the instructions, causing the device to perform operations implementing the MIMO scattering channel estimation method as described in Embodiment 1 or 2 above.

[0131] In summary, this invention utilizes a DT denoising model to perform back-diffusion denoising on noisy scattering channel matrices, effectively suppressing noise interference. By learning the noise distribution characteristics, the DT denoising model gradually recovers the original features of the channel matrix during the back-diffusion denoising process. Compared to traditional estimation methods, this significantly reduces the normalized mean square error of channel estimation and improves estimation accuracy. Furthermore, this invention simulates the forward diffusion process of the channel matrix using a tropospheric scattering channel model, gradually adding Gaussian noise until the channel matrix completely evolves into pure noise. This allows the DT denoising model to learn the complete denoising path from pure noise to a clean channel matrix, thus exhibiting stronger noise resistance in practical applications, especially suitable for low signal-to-noise ratio (SNR) scenarios.

[0132] This invention also utilizes a tropospheric scattering channel model to generate training and validation datasets, which can fully simulate complex channel characteristics such as multipath scattering and time-varying fading in actual communication. By recording the channel state matrix at different time steps, the DT denoising model can learn the time-varying evolution law of the channel matrix, thereby better adapting to the dynamically changing scattering channel environment. By adding Gaussian white noise to the actual pilot signal to generate the test dataset, it more closely resembles the real communication scenario. Combined with the inverse denoising capability of the DT denoising model, this invention can be directly applied to broadband wireless communication systems such as 5G / 6G, improving the performance of MIMO technology in actual deployments and providing technical support for high-reliability, low-latency communication.

[0133] This invention constructs a spatially correlated tropospheric scattering channel model and accurately simulates the multipath scattering characteristics and inter-antenna element correlation of the actual channel based on tapped delay lines and exponential correlation models. The training dataset can fully cover the diffusion process of the channel matrix from a clean state to pure noise. Combined with the Transformer architecture DT denoising model, the feature representation capability is enhanced through real-imaginary part separation processing and temporal position encoding, so that the inverse denoising process can accurately learn the mapping relationship between noise distribution and channel features.

[0134] This invention achieves automatic learning of the joint distribution of noise scaling factor and channel state by iteratively optimizing the loss function. This breaks through the dependence of traditional methods on explicit noise statistical characteristics. In the test dataset, even in the face of pilot contamination or channel correlation mismatch scenarios, it can still quickly restore channel capacity and significantly improve the deployment robustness of 5G / 6G massive MIMO systems in complex scattering environments.

[0135] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0136] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0137] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0138] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0139] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A MIMO scattering channel estimation method, characterized in that, include: Obtain the noisy scattering channel matrix; Based on the noisy scattering channel matrix, back diffusion denoising is performed using the trained DT denoising model to output the denoised scattering channel matrix. The training method for the DT denoising model includes: The scattering channel matrix is ​​forward diffused using a tropospheric scattering channel model, and standard Gaussian noise is gradually added until the scattering channel matrix evolves into pure noise. Training and validation datasets are established by recording the channel state matrix corresponding to the current time step. The received signal is obtained by adding Gaussian white noise to the transmitted pilot signal. The noisy scattering channel matrix is ​​obtained by using the least squares method based on the received signal and the transmitted pilot signal, and a test dataset is established. Using training, validation, and test datasets, the DT denoising model is trained through an iterative method and a reverse denoising process to output the denoised scattering channel matrix. The DT denoising model is trained using an iterative method and a reverse denoising process to output the denoised scattering channel matrix, including: Determine the starting time step of the reverse denoising process; The diffusion signal-to-noise ratio at the current time step is calculated based on the noise scaling factor accumulated at the current time step. Based on the diffusion signal-to-noise ratio of the initial time step and the diffusion signal-to-noise ratio of the current time step, determine the time step that is closest to the diffusion signal-to-noise ratio; The feature enhancement vector of the noisy scattering channel matrix after adding channel state labels is used as the initial value for the reverse denoising process and input into the DT denoising model. Using the DT denoising model, starting from the closest time step, noise estimation is performed step by step on the noisy scattering channel matrix based on the feature enhancement vector to obtain the predicted noise. A loss function is defined with the normalized mean square error between the real noise and the predicted noise as the optimization objective. Based on the loss function, the back-diffusion denoising process is used to iteratively remove noise from the noisy scattering channel matrix, and the denoised scattering channel matrix is ​​output.

2. The MIMO scattering channel estimation method according to claim 1, characterized in that, The scattering channel matrix is ​​forward diffused using a tropospheric scattering channel model, and standard Gaussian noise is gradually added until the scattering channel matrix evolves into pure noise. Training and validation datasets are established by recording the channel state matrix corresponding to the current time step, including: use The tapped channel model architecture is constructed by coupling a tapped delay line structure with an independent Rayleigh fading generator to build a spatially uncorrelated tropospheric scattering channel model. Based on the spatially uncorrelated tropospheric scattering channel model, an exponential correlation model is introduced to simulate the correlation between transmit and receive antenna array elements to construct a spatially correlated tropospheric scattering channel model. Using the forward diffusion process of the DT denoising model, standard Gaussian noise is gradually added to the scattering channel data in the tropospheric scattering channel model with spatial correlation, and the channel state matrix corresponding to the current time step is recorded to obtain the noisy channel data group. Based on the noisy channel data set, training and validation datasets are established by dividing the channel state matrix according to time steps.

3. The MIMO scattering channel estimation method according to claim 2, characterized in that, The spatially uncorrelated tropospheric scattering channel model is expressed as follows: ； In the formula, This represents a spatially uncorrelated tropospheric scattering channel model. , Indicates all dimensions as The space of complex matrices, This indicates the number of transmit antennas at the transmitter T. Indicates the receiving end The number of receiving antennas, Indicates the first Channel matrix of scattering paths, Indicates the first Power of each scattering path Napier's constant is represented by , and L represents the number of scattering paths. Indicates the carrier frequency. Indicates the first The time delay of each scattering path; The spatially correlated tropospheric scattering channel model is expressed as follows: ； In the formula, This represents a spatially correlated tropospheric scattering channel model. and They represent the sending end. The correlation coefficient matrix between the receiver and the receiver; The formula for generating the channel state matrix corresponding to the current time step is expressed as: ； In the formula, Indicates the current time step The corresponding channel state matrix, Indicates the current time step Cumulative noise scaling factor Indicates the current time step The corresponding channel state matrix at that time, Indicates standard Gaussian noise. Indicates the current time step The noise scaling factor, Indicates the current time step noise factor Indicates from arrive The product of two digits.

4. The MIMO scattering channel estimation method according to claim 1, characterized in that, The received signal is obtained by adding Gaussian white noise to the transmitted pilot signal. The noisy scattering channel matrix is ​​then obtained using the least squares method based on the received and transmitted pilot signals. A test dataset is established, including: Gaussian white noise is added to the transmitted pilot signal to obtain the received signal; The scattering channel matrix is ​​decoupled from the transmitted pilot signal using the least squares method to obtain the noisy scattering channel matrix; A test dataset was constructed using the noisy scattering channel matrix as the source data and the channel state matrix as the label.

5. The MIMO scattering channel estimation method according to claim 4, characterized in that, The received signal is represented as: ； In the formula, Indicates receiving signal, This represents a spatially correlated tropospheric scattering channel model. Indicates the transmission of pilot signals. And satisfy A denotes the conjugate transpose of a matrix. Indicates all dimensions as The space of complex matrices, Indicates the sending end The number of transmitting antennas, Indicates the length of the transmitted pilot. Let N denote the identity matrix, and let N denote additive white Gaussian noise. The source data is represented as follows: ； In the formula, Indicates source data, This represents the conjugate transpose of the pilot signal.

6. The MIMO scattering channel estimation method according to claim 1, characterized in that, The DT denoising model is constructed based on the Transformer architecture. Before performing back-diffusion denoising processing based on the trained DT denoising model according to the noisy scattering channel matrix, the following steps are also included: The noisy scattering channel matrix is ​​divided into two channels, real and imaginary, and then transformed into a high-dimensional feature vector through linear embedding. The time step corresponding to the noisy scattering channel matrix is ​​converted into a set of time position codes using the cofactor selection function; The high-dimensional feature vector is added to the temporal location encoding to form the feature enhancement vector, which is then input into the trained DT denoising model.

7. The MIMO scattering channel estimation method according to claim 1, characterized in that, The loss function is expressed as: ； in, Represents the loss function. For the network parameters of the Transformer architecture, Indicates the current time step Current time step Channel state matrix and current time step Real noise Expected value Indicates the current time step Channel state matrix and current time step The noise estimation function, This represents the square of the L2 norm.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the MIMO scattering channel estimation method as described in any one of claims 1-7.

9. A computer device, characterized in that, include: Memory, used to store instructions; A processor for executing the instructions, causing the device to perform operations implementing the MIMO scattering channel estimation method as described in any one of claims 1-7.

Images