RIS-assisted V2X channel estimation method

By combining RIS and deep convolution in V2X channels to generate adversarial networks, the channel estimation method is optimized, and the problem of insufficient channel estimation accuracy is solved, improving the accuracy and system performance of channel estimation.

CN119030832APending Publication Date: 2024-11-26SHANGHAI INST OF TECH
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411273659.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-11
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

In the complex V2X channel environment, traditional channel estimation methods are difficult to accurately track channel changes, resulting in insufficient channel estimation accuracy, affecting the reliability of Internet of Vehicles communication and system scalability.

Method used

Using an intelligent metasurface (RIS)-assisted method, the compression perception method is combined with the deep convolution generation adversarial network, and the coarse estimation results of the compression perception algorithm are optimized through the deep convolution generation adversarial network model to improve the accuracy of channel estimation.

Benefits of technology

It significantly improves the accuracy and accuracy of V2X channel estimation, reduces the complexity of channel estimation, and enhances the reliability and system performance of Internet of Vehicles communication.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119030832A_ABST
    Figure CN119030832A_ABST
Patent Text Reader

Abstract

The invention relates to the related field of a wireless communication channel estimation technology, and discloses an intelligent metasurface-assisted channel estimation method in an Internet of Vehicles environment. The method comprises the following steps: converting a channel estimation problem into a compressed sensing reconstruction problem by using the sparsity of millimeter waves, obtaining an initial estimation value of a cascade channel by using a compressed sensing algorithm, and carrying out noise reduction on a channel matrix subjected to initial estimation by using a deep convolution generative adversarial model in order to further improve the accuracy of channel estimation in an Internet of Vehicles environment so as to improve the accuracy of channel estimation in an Internet of Vehicles environment. And the estimation precision of the channel estimation method is improved. In the field of Internet of Vehicles communication, more reliable guarantee and higher traffic efficiency can be provided for vehicle communication and road safety.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of wireless communication technologies, and in particular to a RIS-assisted V2X channel estimation method. Background Art

[0002] With the commercialization of the fifth generation of mobile communication technology and its gradual extension to various fields, Vehicle-to-Everything (V2X), as an important branch of the Internet of Things in the transportation system, has been greatly affected. Some advanced communication network technologies have been applied to V2X. However, there are still many problems in the complex environment of V2X mobile communication.

[0003] As vehicles move at high speeds, channel characteristics (such as fading, delay spread, etc.) change rapidly, making it difficult for traditional channel estimation methods to accurately track them. In addition, obstacles such as buildings and trees in urban environments can cause reflection, scattering, and diffraction of signal propagation, making channel estimation more complex and uncertain. Internet of Vehicles applications (such as autonomous driving and traffic safety systems) have extremely high requirements for communication reliability, and errors in channel estimation can lead to serious consequences.

[0004] Existing technologies face a lot of challenges, the most typical of which include high bandwidth requirements, high-speed mobility of vehicles, system scalability, and communication reliability. Since the actual vehicle channel is very complex and V2X has high requirements for signal processing speed and accuracy, further reducing the complexity of the channel estimation method and improving its accuracy is crucial for the further development of V2X. Summary of the invention

[0005] In view of the problem of insufficient estimation accuracy of typical channel estimation methods in existing vehicle networks, the present invention proposes a V2X channel estimation method assisted by a Reconfigurable Intelligent Surface (RIS). By utilizing the sparse characteristics of the V2X channel of the RIS-assisted wireless communication system, the compressed sensing method is combined with a deep generative adversarial network. The deep convolutional generative adversarial network model is used to further optimize the rough estimation results of the compressed sensing algorithm, so as to improve the accuracy of channel estimation in complex V2X environments.

[0006] In order to achieve the above-mentioned invention object, the present invention adopts the following technical scheme:

[0007] The present invention provides a RIS-assisted V2X channel estimation method, which is characterized by: utilizing the sparse characteristics of the V2X channel of the RIS-assisted wireless communication system, combining the compressed sensing method with a deep convolutional generative adversarial network, and using the deep convolutional generative adversarial network model to further optimize the rough estimation result of the compressed sensing algorithm.

[0008] Furthermore, the present invention provides a RIS-assisted V2X channel estimation method, characterized in that it includes the following steps:

[0009] S1: Construct a wireless communication system model in the Internet of Vehicles environment;

[0010] S2: Convert the cascade channel model into an angle domain cascade channel and write a compressed sensing model based on the received signal;

[0011] S3: According to the sparsity of the cascaded channel, the channel estimation problem is transformed into a compressed sensing recovery problem, and the sparsity adaptive matching pursuit algorithm is used to obtain a rough channel estimation matrix;

[0012] S4: Introduce deep learning methods in RIS-assisted V2X cascade channel estimation, and use a deep convolutional generative adversarial network model to optimize and reduce the noise of the channel matrix to obtain more accurate channel estimation results.

[0013] Furthermore, the present invention provides a RIS-assisted V2X channel estimation method, characterized in that:

[0014] In S1, a wireless communication system model in a connected vehicle environment is established based on the millimeter wave channel model.

[0015] Furthermore, the present invention provides a RIS-assisted V2X channel estimation method, characterized in that:

[0016] In S1, the following specific steps are included:

[0017] The user sends the pilot symbol of known information to the base station through RIS assistance. At time t, the pilot sent by the kth user, the received signal at the base station is It is expressed as:

[0018] y k (t) = Gθu k (t)s k (t)+w(t)

[0019] in, Indicates BS-RIS channel, represents the channel from RIS to the kth terminal, s k (t) is the pilot signal sent by the kth user, w(t) is the additive white Gaussian noise;

[0020] BS controls the unit controller on RIS to adjust the phase shift and amplitude of each element. The phase shift matrix of RIS It is expressed as:

[0021]

[0022] Where diag(·) represents a diagonal matrix; ω m ∈[0,2π);ω m describes the phase shift of the RIS, β m ∈[0,1],m∈(1,2,…,M),β m describes the scaling of amplitude and power loss;

[0023] RIS consists of M reflection units and uses a UPA antenna array. According to the response of the UPA antenna array, the RIS array response vector is defined as:

[0024]

[0025] Where M = M1 × M2, d is the antenna spacing, λ represents the wavelength, and d is the antenna spacing;

[0026] Channel G of BS-RIS and channel u of RIS-UE k Respectively expressed as:

[0027]

[0028] in:(·) H represents the conjugate transpose, L G and L r,k denote the number of paths from BS to RIS and the number of paths from RIS to the kth terminal respectively; εl and are the complex gains composed of path losses in BS-RIS and RIS-UE channels respectively; and Indicates BS azimuth and elevation; and Indicates the azimuth and elevation of RIS; and are the azimuth and elevation of RIS, and denote the antenna response vectors of the two UPAs at BS and RIS respectively;

[0029] Cascaded channel matrix of BS-RIS and RIS-UE It is expressed as H = Gdiag (u k )

[0030]

[0031] So we can get the received signal

[0032]

[0033] Among them, the RIS reflection coefficient Anddiag(θ(t))uk =diag(u k )θ(t),

[0034] After T time instants, the received signal at the base station is sent by the kth user. for

[0035] Y k =HΦ+W K

[0036] Among them, Φ=[θ(1), θ(2),…, θ(T)], W K =[W k (1), w k (2),…,w k (T)].

[0037] Furthermore, the present invention provides a RIS-assisted V2X channel estimation method, characterized in that:

[0038] In S2, the cascade channel H is represented as:

[0039] H=G H diag(u k )

[0040] in, Indicates BS-RIS channel, represents the channel from RIS to the kth terminal;

[0041] In the angle domain, the cascaded channel H is expressed as:

[0042]

[0043] in, is the concatenated channel representation in the angle domain, and is a dictionary unitary matrix;

[0044] The expression of the received signal matrix of the kth user is:

[0045]

[0046] Among them, Y k represents the effective received signal from the kth user to the BS, which can be regarded as an effective measurement value matrix, Φ is the reflection coefficient matrix of RIS, and W k is the noise matrix;

[0047] The compressed sensing model is:

[0048] Furthermore, the present invention provides a RIS-assisted V2X channel estimation method, characterized in that:

[0049] In S3, by inputting the measurement matrix Sensing Matrix Ψ i , set the threshold ε, the maximum number of iterations T and the support set Λ0, and initialize the signal residual r0 and the support set Λ0;

[0050] Calculate the inner product of the columns of the measurement matrix and the current residual to obtain the matching degree of each column with the residual;

[0051] Select the column with the largest match and add it to the index set Λ0;

[0052] updating the signal estimate and the signal residual according to the selected columns and the measurement matrix;

[0053] Find the least squares solution If the threshold is reached or the maximum number of iterations is reached, the iteration is stopped and the output is

[0054] Finally, the channel matrix is ​​transformed into the spatial domain to obtain a rough value of the channel estimate, that is,

[0055] Furthermore, the present invention provides a RIS-assisted V2X channel estimation method, characterized in that:

[0056] In S4, the deep convolutional generative adversarial network model models the channel estimation problem as a restoration problem from a low-resolution image to a high-resolution image, and roughly estimates the channel matrix Perform noise reduction to obtain a more accurate channel estimation value H real , utilizing the sparsity of the channel in massive MIMO systems, the channel matrix is ​​regarded as a two-dimensional image, and channel estimation is performed by converting it into an image processing method.

[0057] Furthermore, the present invention provides a RIS-assisted V2X channel estimation method, characterized in that:

[0058] In S4, the purpose of constructing the DCGAN model is to use limited data to learn the probability distribution of all data as much as possible. It introduces deep convolutional neural networks in the generator and discriminator of the generative adversarial network model to increase the stability of the network and generate higher quality samples. The generator is responsible for converting the input noisy channel matrix into a more accurate channel matrix, while the discriminator is responsible for judging whether the generated channel matrix is ​​realistic. Through continuous adversarial training of the two networks, the generator can eventually generate higher quality samples.

[0059] In addition, the present invention also provides a channel estimation device / system for RIS-assisted cascade channels, which is characterized by comprising an initialization module, a channel sensing data acquisition module, a compressed sensing estimation processing module, a noise reduction module, and a channel estimation application module;

[0060] The initialization module includes a transmitter, a receiver and a RIS reflection unit, constructs a communication system in a vehicle networking environment, and initializes the configuration of the RIS reflection unit;

[0061] The channel sensing data acquisition module sends a known training sequence from the transmitting end to the receiving end, adjusts different reflection modes through the RIS reflection unit to receive the signal, and the receiving end records the received signal and obtains the sensing data in combination with the known training sequence of the transmitting end;

[0062] The compressed sensing algorithm estimation processing module processes the received signal, uses the compressed sensing algorithm to perform channel estimation, constructs a sparse representation model of the channel, and uses the compressed sensing algorithm to solve a rough channel estimation matrix;

[0063] The above-mentioned denoising module constructs a DCGAN model, including a generator and a discriminator. The rough channel matrix obtained by the compressed sensing algorithm is input into the generator. The generator generates the denoised channel matrix, and the discriminator makes a true or false judgment. Through adversarial training, the parameters of the generator and the discriminator are optimized, so that the generator can output a high-quality channel matrix and finally output a more accurate channel estimation value.

[0064] The above channel estimation result application module applies the accurate channel estimation value to the Internet of Vehicles communication system to optimize the transmission performance of the communication system, dynamically adjusts the configuration of the RIS reflection unit according to the estimated channel matrix, and improves the channel capacity and the overall performance of the system.

[0065] In addition, the present invention also provides a carrier, characterized in that: it includes a memory and a processor;

[0066] The memory stores a computer program that can be run on the processor;

[0067] When the processor executes the computer program, any of the channel estimation methods for the RIS-assisted V2X environment is implemented.

[0068] Compared with the prior art, the present invention has the following advantages:

[0069] The present invention adopts the SAMP algorithm to first roughly estimate the RIS-assisted cascade channel, uses a deep convolutional generative adversarial network for training, and then performs noise reduction on the roughly estimated channel, thereby making the channel prediction more realistic and accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 It is a schematic diagram of the structure of the RIS-assisted communication system model of the present invention.

[0071] Figure 2 It is a flow chart of the RIS-assisted V2X channel estimation method in an embodiment of the present invention.

[0072] Figure 3 This is a DNGAN network model diagram in an embodiment of the present invention.

[0073] Figure 4 is a diagram of a channel estimation system in an embodiment of the present invention. DETAILED DESCRIPTION

[0074] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention more clear, the present invention is described in detail below with reference to the accompanying drawings.

[0075] Embodiment 1:

[0076] The main research object of this paper is V2X channel estimation, assuming that all channels have no obstruction or other unfavorable propagation environment. Figure 1 The RIS-assisted multi-terminal MIMO system in the time division duplex (TDD) mode shown in the figure, wherein the channel is affected by block fading, that is, within a coherence time, the channel parameters remain unchanged. A coherence time is divided into two parts, including uplink channel estimation and downlink information transmission. A RIS is deployed between the BS and the user to assist in information transmission, and in order to better study the estimation problem of the RIS-assisted wireless channel, the present invention only considers the link reflected by the RIS, and ignores the direct transmission link from the BS to the UE. The RIS-assisted V2X channel estimation method process in the present invention is as follows: Figure 2 shown.

[0077] Step 1: Establish a wireless communication system model in the Internet of Vehicles environment based on the millimeter wave channel model. Specifically, the user sends a pilot symbol of known information to the base station through the assistance of RIS. At time t, the pilot sent by the kth user, the received signal at the base station is 1 can be expressed as:

[0078] y k (t) = Gθu k (t)s k (t)+w(t)

[0079] in, Indicates BS-RIS channel, represents the channel from RIS to the kth terminal, s k(t) is the pilot signal sent by the kth user, and w(t) is additive white Gaussian noise. Using the orthogonal pilot transmission strategy, assuming that the pilot signal s k =1.

[0080] BS can control the unit controller on RIS through wired or wireless connection to adjust the phase shift and amplitude of each element. The phase shift matrix of RIS It can be expressed as:

[0081]

[0082] Where: diag(·) represents a diagonal matrix; ω m ∈[0,2π);ω m describes the phase shift of the RIS, β m ∈[0,1],m∈(1,2,…,M),β m describes the amplitude scaling and power loss, but RIS cannot achieve complete reflection in practice. Assuming β m = 1. RIS consists of M reflection units and uses a UPA antenna array. According to the response of the UPA antenna array, the RIS array response vector is defined as:

[0083]

[0084] Where M = M1 × M2, d is the antenna spacing, λ represents the wavelength, and d is the antenna spacing, which is usually taken as d = λ / 2.

[0085] Since the scattering path of the millimeter wave channel is limited, its channel model has rich geometric features, which can be used to convert the BS-RIS channel G and the RIS-UE channel u k Respectively expressed as:

[0086]

[0087] in:(·) H represents the conjugate transpose, L G and L r,k They represent the number of paths from BS to RIS and the number of paths from RIS to the kth terminal respectively; ρ l and are the complex gains composed of path losses in BS-RIS and RIS-UE channels respectively; and Indicates BS azimuth and elevation; and Indicates the azimuth and elevation of RIS; and are the azimuth and elevation of RIS. and Denote the antenna response vectors of the two UPAs at the BS and RIS respectively.

[0088] Cascaded channel matrix of BS-RIS and RIS-UE It can be expressed as

[0089] H=Gdiag(u k )(11)

[0090]

[0091] So we can get the received signal

[0092]

[0093] Among them, the RIS reflection coefficient Anddiag(θ(t))u k =diag(u k )θ(t).

[0094] After T time instants, the received signal at the base station is sent by the kth user. Can write

[0095] Y k =HΦ+W K

[0096] Among them, Φ=[θ(1), θ(2),…, θ(T)], W K =[w k (1), w k (2),…,w k (T)].

[0097] Step 2: Convert the cascade channel model into an angle domain cascade channel and write a compressed sensing model based on the received signal.

[0098] RIS-assisted V2X channel estimation method, the cascade channel H can be expressed as:

[0099] H=G H diag(u k )

[0100] in, Indicates BS-RIS channel, represents the channel from RIS to the kth terminal.

[0101] In the angle domain, the cascade channel H can be expressed as:

[0102]

[0103] in, is the concatenated channel representation in the angle domain, and is a dictionary unitary matrix.

[0104] Then the expression of the received signal matrix of the kth user is:

[0105]

[0106] Where Yk represents the effective received signal from the kth user to the BS, which can be regarded as an effective measurement value matrix, Φ is the reflection coefficient matrix of RIS, and W k is the noise matrix. The above formula is written as a compressed sensing model:

[0107]

[0108] Step 3: Cascade channels It is sparse. Due to the high frequency of millimeter waves, it will experience greater attenuation and blocking during transmission, which makes the millimeter wave channel present a sparse characteristic in the angle domain. Taking advantage of this sparse characteristic, the angle domain cascade channel estimation problem can be converted into a sparse matrix recovery problem by converting the channel to be estimated from the spatial domain to the virtual angle domain of angle quantization. The SAMP algorithm can provide efficient channel estimation in sparse environments by gradually increasing the sparsity and gradually optimizing the channel estimation. The SAMP algorithm is used to recover the sparse channel matrix, thereby converting the RIS-assisted V2X cascade channel estimation problem into a known effective measurement matrix. and sensor matrix Recovering the angular domain cascade channel problem.

[0109] Firstly, the channel estimation problem is transformed into a compressed sensing recovery problem according to the sparsity of the cascaded channel. Then, the SAMP algorithm is used to obtain a rough channel estimation matrix.

[0110] Bring in the angle domain cascade channel and receive the signal Y k It can be expressed as:

[0111]

[0112] Further simplification yields:

[0113]

[0114] The channel to be estimated is converted from the spatial domain to the virtual angular domain, and the sparsity is used to transform the RIS-assisted cascade channel estimation problem into a known effective measurement matrix. and sensor matrix Recovering the angular domain cascade channel problem.

[0115] The basic idea of ​​the SAMP algorithm is to gradually approximate the sparse representation of the signal through iteration. In each iteration, the algorithm selects the column most relevant to the current residual based on the current residual and measurement matrix, and updates the estimated values ​​of the residual and signal. Through continuous iteration, signal reconstruction is finally achieved.

[0116] The algorithm steps are as follows: By inputting the measurement matrix Sensing Matrix Ψ i , set the threshold ε, the maximum number of iterations T and the support set Λ0. Initialize the signal residual r0 and the support set Λ0. Calculate the inner product of the columns of the measurement matrix and the current residual to obtain the matching degree of each column with the residual. Select the column with the largest matching degree and add it to the index set Λ0. Update the signal estimate and signal residual according to the selected columns and measurement matrix. Find the least squares solution If the threshold is reached or the maximum number of iterations is reached, the iteration is stopped and the output is Finally, the channel matrix is ​​transformed into the spatial domain, that is, Get a rough value of the channel estimate.

[0117] Step 4: To further improve the estimation accuracy, a deep learning method is introduced into the PIS-assisted V2X cascade channel estimation, and a deep convolutional generative adversarial network model is used to optimize and reduce the noise of the channel matrix to obtain a more accurate channel estimation result.

[0118] Based on the rough value of the RIS-assisted V2X cascade channel matrix estimated by the compressed sensing algorithm, a deep convolution model based on the generative adversarial network is used to reduce the noise of the initially estimated channel matrix to improve the estimation accuracy of the channel estimation method. The details are as follows:

[0119] When estimating the cascade channel, the estimated value of the SAMP algorithm is used as the input data of the deep convolutional model of the generative adversarial network. Based on the above step 3, the channel estimated value estimated by the SAMP algorithm can be written as

[0120]

[0121] We further designed a deep convolutional generative adversarial network model to model the channel estimation problem as a restoration problem from a low-resolution image to a high-resolution image, and roughly estimated the channel matrix Perform noise reduction to obtain a more accurate channel estimation value H real Taking advantage of the sparsity of the channel in massive MIMO systems, the channel matrix is ​​regarded as a two-dimensional image and channel estimation is performed by converting it into an image processing method.

[0122] Generative adversarial networks are a type of deep learning model. GANs consist of two different models: a generator and a discriminator. The structure of a deep convolutional generative adversarial network is similar to that of a GAN network. A generative adversarial network may use fully connected layers or simple multi-layer perceptrons (MLPs) as the network structure of the generator and discriminator, which may not work well when processing high-dimensional image data. The network structure of the generator and discriminator in DCGAN is usually based on convolutional layers rather than fully connected layers. This makes DCGAN more suitable for processing image data and can better capture the spatial structure and local features of images.

[0123] Specifically, a rough noisy channel estimation matrix is ​​received as the input of the generator network, and then a flat convolution layer, four down convolution blocks, and one convolution layer per convolution block are used to extract feature attributes from the input image. Batch normalization and LReLU activation function are added to each convolution layer to accelerate the learning process and enhance the expressiveness of the model. The LReLU activation function can be expressed as:

[0124]

[0125] Collect real channel matrix data for training the discriminator. Set up sensors and measuring equipment in the Internet of Vehicles environment to obtain channel state information between the transmitter and the receiver. During the acquisition process, record channel data at different times, locations, and moving speeds. Use simulation tools to set parameters according to specific scenarios to generate channel matrices under different conditions. Preprocess the channel matrix obtained by acquisition or simulation, including normalization, noise reduction, and filling in missing data. Divide the processed channel matrix into training and test sets. Use the processed real channel matrix as one of the real inputs of the discriminator. During the training process, the real channel matrix needs to be used as a reference standard for the discriminator. The goal of training is to make the channel matrix generated by the generator as close to the real channel matrix as possible. However, in practical applications, the real channel matrix is ​​usually not directly available.

[0126] The channel matrix generated by the generator and the true channel matrix are received as input. The discriminant network uses a convolution layer with a step size of 2 instead of the pooling layer. On the one hand, it can reduce the amount of calculation of the model, and on the other hand, the pooling layer will cause some information loss. The input of the discriminator is the preliminary estimated channel estimation matrix and the output of the generator, that is, the input of the discriminant network is the reference image and the depth image generated by the generative network. The discriminant network uses 6 convolutional layers to learn the features of the image to distinguish between true and false. Batch normalization and Leaky ReLU activation functions are added to each of the first five convolutional layers to accelerate convergence and control overfitting. In the last layer of the discriminator, the Sigmoid activation function is used for feature mapping for probability analysis, and the discrimination results are normalized to [0, 1]. The Sigmoid activation function can be expressed as:

[0127]

[0128] The purpose of building a DCGAN model is to use limited data to learn the probability distribution of all data as much as possible. It introduces deep convolutional neural networks in the generator and discriminator of the GAN model to increase the stability of the network and generate higher quality samples.

[0129] The discriminator uses the real channel matrix and the channel matrix generated by the generator to calculate the classification error between the two and update the discriminator parameters to more accurately distinguish the real and generated channel matrices. Through the channel matrix generated by the generator, the results of the discriminator output are back-propagated to the generator, and the generator parameters are updated to make the channel matrix generated by the generator closer to the real channel matrix.

[0130] The structure of the deep convolutional generative adversarial network is similar to that of the GAN network. A deep convolutional neural network is introduced into the generator and the discriminator. Its optimization objective function can be expressed as:

[0131]

[0132] Among them, H real is the real sample image, i.e. the real channel, is the noisy and distorted image input into the network, and G(z) is the generated image, that is, the generated channel matrix.

[0133] The loss function is used to evaluate the inconsistency between the true value and the predicted value of the model. The smaller the loss function, the better the robustness of the model. By observing the loss function of the generator and the discriminator, when the loss function reaches a certain stable value or no longer changes significantly, it can be considered that the model has converged. DCGAN defines the loss function of the generator L G And the loss function L of the discriminator D for

[0134]

[0135] Where z is the noise, is the random noise distribution; G(z) is the sample generated by the generator; D(G(z)) is the probability that the discriminator identifies the generated sample as a real sample; is the true sample distribution; D(x) is the probability that the discriminator identifies the true sample as a true sample.

[0136] In DCGAN, both the discriminant network and the generative network use convolutional neural networks instead of the traditional GAN ​​multi-layer perception. The DCGAN network model is as follows: Figure 3As shown, the model is mainly composed of a generator and a discriminator. During the training process, the generator and the discriminator are trained alternately, and adversarial learning is achieved by minimizing the loss of the generator and the discriminator. The generator gradually extracts and generates more refined features through multiple convolutional layers and upsampling layers, and finally outputs a denoised channel estimation matrix. The discriminator receives the real channel estimation matrix and the channel estimation matrix generated by the generator, gradually reduces the size of the feature map through convolution and downsampling, and finally outputs a binary value, indicating whether the input is real or generated. Based on the present invention, the channel is first preprocessed by a compressed sensing algorithm to achieve a preliminary estimation of the channel, and then the preprocessed signal is regarded as a dual-channel image as a conditional input. The two channels of the image represent the real part and the imaginary part of the complex matrix respectively, so that the channel estimation problem can be converted into a problem of restoring a high-resolution image with a given low-resolution image; then the received signal is first preprocessed by a compressed sensing algorithm estimator, and then the output is sent to the trained model to achieve accurate channel estimation.

[0137] Embodiment 2:

[0138] On the other hand, the embodiment of the present invention also provides a channel estimation device and system for a RIS-assisted cascade channel, such as Figure 4 It includes: initialization module, channel sensing data acquisition module, compressed sensing estimation processing module, noise reduction module, and channel estimation application module.

[0139] System initialization module: Builds the communication system in the Internet of Vehicles environment, including the transmitter, receiver, and RIS reflection unit. Initializes the configuration of the RIS reflection unit to ensure that the signal can be effectively reflected.

[0140] Channel sensing data acquisition module: The transmitter sends a known training sequence to the receiver, and adjusts different reflection modes through the RIS reflection unit to receive the signal. The receiver records the received signal and combines it with the known training sequence of the transmitter to obtain the sensing data.

[0141] Compressed sensing algorithm estimation processing module: processes the received signal, uses the compressed sensing algorithm to perform channel estimation, builds a sparse representation model of the channel, and uses the compressed sensing algorithm to solve a rough channel estimation matrix.

[0142] Denoising module: Build a DCGAN model, including a generator and a discriminator. The rough channel matrix obtained by the compressed sensing algorithm is used as input to the generator. The generator generates a denoised channel matrix, and the discriminator makes a true or false judgment. Through adversarial training, the parameters of the generator and the discriminator are optimized so that the generator can output a high-quality channel matrix. Finally, a more accurate channel estimate is output.

[0143] Channel estimation result application module: Apply accurate channel estimation values ​​to the Internet of Vehicles communication system to optimize the transmission performance of the communication system. According to the estimated channel matrix, dynamically adjust the configuration of the RIS reflection unit to improve the channel capacity and the overall performance of the system.

[0144] On the other hand, an embodiment of the present invention further provides a chip, including a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the computer program, any of the above-mentioned channel estimation methods for the RIS-assisted V2X environment is implemented.

[0145] The above describes the specific embodiments of the present invention. The present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention.

Claims

1. A RIS-assisted V2X channel estimation method, characterized in that: Taking advantage of the sparse characteristics of the V2X channel of the RIS-assisted wireless communication system, the compressed sensing method is combined with the deep convolutional generative adversarial network, and the deep convolutional generative adversarial network model is used to further optimize the rough estimation results of the compressed sensing algorithm.

2. The RIS-assisted V2X channel estimation method according to claim 1, characterized in that: The following steps are involved: S1: Construct a wireless communication system model in the Internet of Vehicles environment; S2: Convert the cascade channel model into an angle domain cascade channel and write a compressed sensing model based on the received signal; S3: According to the sparsity of the cascaded channel, the channel estimation problem is transformed into a compressed sensing recovery problem, and the sparsity adaptive matching pursuit algorithm is used to obtain a rough channel estimation matrix; S4: Introduce deep learning methods in RIS-assisted V2X cascade channel estimation, and use a deep convolutional generative adversarial network model to optimize and reduce the noise of the channel matrix to obtain more accurate channel estimation results.

3. The RIS-assisted V2X channel estimation method according to claim 1, characterized in that: In S1, a wireless communication system model in a connected vehicle environment is established based on the millimeter wave channel model.

4. The RIS-assisted V2X channel estimation method according to claim 1, characterized in that: In S1, the following specific steps are included: The user sends the pilot symbol of known information to the base station through the assistance of RIS. At time t, the pilot sent by the kth user, the received signal at the base station is It is expressed as: in, Indicates BS-RIS channel, represents the channel from RIS to the kth terminal, s k (t) is the pilot signal sent by the kth user, w(t) is the additive white Gaussian noise; BS controls the unit controller on RIS to adjust the phase shift and amplitude of each element. The phase shift matrix of RIS It is expressed as: Where diag(·) represents a diagonal matrix; ω m ∈[0,2π);ω m describes the phase shift of the RIS, β m ∈[0,1],m∈(1,2,…,M),β m describes the scaling of amplitude and power loss; RIS consists of M reflection units and uses a UPA antenna array. According to the response of the UPA antenna array, the RIS array response vector is defined as: Where M = M1 × M2, d is the antenna spacing, λ represents the wavelength, and d is the antenna spacing; Channel G of BS-RIS and channel u of RIS-UE k Respectively expressed as: in:(·) H represents the conjugate transpose, L G and L r , k represent the number of paths from BS to RIS and the number of paths from RIS to the kth terminal respectively; ρ l and are the complex gains composed of path losses in BS-RIS and RIS-UE channels respectively; and Indicates BS azimuth and elevation; and Indicates the azimuth and elevation of RIS; and are the azimuth and elevation of RIS, and denote the antenna response vectors of the two UPAs at BS and RIS respectively; Cascaded channel matrix of BS-RIS and RIS-UE It is expressed as H = Gdiag (u k ) So we can get the received signal Among them, the RIS reflection coefficient θ=e jw , and diag(θ(t))u k =diag(u k )θ(t), After T time instants, the received signal at the base station is sent by the kth user. for Y k =HΦ+W K Where, Φ = [θ(1), θ(2),..., θ(T)], W K =[w k (1),w k (2),...,w k (T)].

5. The RIS-assisted V2X channel estimation method according to claim 1, characterized in that: In S2, the cascade channel H is represented as: H=G H god(u k ) in, Indicates BS-RIS channel, represents the channel from RIS to the kth terminal; In the angle domain, the cascaded channel H is expressed as: in, is the concatenated channel representation in the angle domain, and is a dictionary unitary matrix; The expression of the received signal matrix of the kth user is: Among them, Y k represents the effective received signal from the kth user to the BS, which can be regarded as an effective measurement value matrix, Φ is the reflection coefficient matrix of RIS, and W k is the noise matrix; The compressed sensing model is:

6. A RIS-assisted V2X channel estimation method as claimed in claim 5, characterized in that: In S3, by inputting the measurement matrix Sensing Matrix Ψ i , set the threshold ε, the maximum number of iterations T and the support set Λ0, and initialize the signal residual r0 and the support set Λ0; Calculate the inner product of the columns of the measurement matrix and the current residual to obtain the matching degree of each column with the residual; Select the column with the largest match and add it to the index set Λ0; updating the signal estimate and the signal residual according to the selected columns and the measurement matrix; Find the least squares solution If the threshold is reached or the maximum number of iterations is reached, the iteration is stopped and the output is Finally, the channel matrix is ​​transformed into the spatial domain to obtain a rough value of the channel estimate, that is, 7. The RIS-assisted V2X channel estimation method according to claim 6, characterized in that: In S4, the deep convolutional generative adversarial network model models the channel estimation problem as a restoration problem from a low-resolution image to a high-resolution image, and roughly estimates the channel matrix Perform noise reduction to obtain a more accurate channel estimation value H real , utilizing the sparsity of the channel in massive MIMO systems, the channel matrix is ​​regarded as a two-dimensional image, and channel estimation is performed by converting it into an image processing method.

8. The RIS-assisted V2X channel estimation method according to claim 7, characterized in that: In S4, the purpose of constructing the DCGAN model is to use limited data to learn the probability distribution of all data as much as possible. It introduces deep convolutional neural networks in the generator and discriminator of the generative adversarial network model to increase the stability of the network and generate higher quality samples. The generator is responsible for converting the input noisy channel matrix into a more accurate channel matrix, while the discriminator is responsible for judging whether the generated channel matrix is ​​realistic. Through continuous adversarial training of the two networks, the generator can eventually generate higher quality samples.

9. A channel estimation device / system for RIS-assisted cascaded channels, characterized in that: It includes an initialization module, a channel sensing data acquisition module, a compressed sensing estimation processing module, a noise reduction module, and a channel estimation application module; The initialization module includes a transmitting end, a receiving end and a RIS reflection unit, constructs a communication system in a vehicle networking environment, and initializes the configuration of the RIS reflection unit; The channel sensing data acquisition module sends a known training sequence from the transmitting end to the receiving end, adjusts different reflection modes through the RIS reflection unit to receive the signal, and the receiving end records the received signal and obtains the sensing data in combination with the known training sequence of the transmitting end; The compressed sensing algorithm estimation processing module processes the received signal, uses the compressed sensing algorithm to perform channel estimation, constructs a sparse representation model of the channel, and uses the compressed sensing algorithm to solve a rough channel estimation matrix; The denoising module constructs a DCGAN model, including a generator and a discriminator. The rough channel matrix obtained by the compressed sensing algorithm is input into the generator. The generator generates a denoised channel matrix, and the discriminator performs authenticity judgment. Through adversarial training, the parameters of the generator and the discriminator are optimized, so that the generator can output a high-quality channel matrix and finally output a more accurate channel estimation value. The channel estimation result application module applies the accurate channel estimation value to the Internet of Vehicles communication system to optimize the transmission performance of the communication system, dynamically adjusts the configuration of the RIS reflection unit according to the estimated channel matrix, and improves the channel capacity and the overall performance of the system.

10. A carrier, characterized in that: including memory and processor; The memory stores a computer program executable on the processor; When the processor executes the computer program, any of the above-mentioned channel estimation methods for a RIS-assisted V2X environment is implemented.

Citation Information

Cited By

  • Indoor multi-target visible light positioning method based on compressed sensing

    CN120028751A