A Channel Estimation Method for Intelligent Metasurface-Assisted High-Speed Railway Wireless Communication

Abstract

The present invention provides a channel estimation method for intelligent metasurface-assisted high-speed rail wireless communication. First, a link channel model is established based on the communication system, and then the least squares algorithm is used for preliminary channel estimation to obtain noisy channel data. Next, a channel estimation network model based on deep learning is constructed, and the noisy channel data is used as the input of the channel estimation network model. In the channel estimation network model, the pilot signal and the channel matrix are used as two-channel images with different sizes. Then, the loss function of the channel estimation network model is optimized, the estimation error is calculated according to the estimated channel and backpropagated to train the model parameters until the model parameters converge, so as to obtain the trained channel estimation network model. Finally, the trained channel estimation network model is used for channel estimation. The present invention can achieve accurate channel estimation results with a small number of pilots.

Description

Technical Field

[0001] The present invention relates to the technical field of wireless communication, and particularly to a channel estimation method for intelligent surface-assisted high-speed rail wireless communication. Background Art

[0002] As an emerging technology to improve the coverage range and resource utilization efficiency of wireless systems, Reconfigurable Intelligent Surface (RIS) has attracted extensive attention from researchers. Introducing RIS into a wireless communication system can make the transmission environment controllable and effectively improve the quality of wireless communication services.

[0003] The good performance gain brought by RIS depends on accurate channel state information. During the communication process, in order to decode the received data as accurately as possible, it is necessary to estimate the characteristics of the channel from the state of the received signal. Therefore, channel estimation is an essential part of the RIS-assisted communication system.

[0004] However, different from public wireless networks, the high-speed rail network has more stringent reliability requirements. In the high-speed rail scenario, not only does the rapid movement of the train lead to frequent handovers, severe Doppler frequency shifts, and smaller channel coherence intervals, but also the communication scenario is diverse and complex. In the RIS-assisted high-speed rail communication scenario, how to accurately estimate the channel is a technical problem that needs to be solved by those skilled in the art. Summary of the Invention

[0005] In view of this, the present invention provides a channel estimation method for intelligent surface-assisted high-speed rail wireless communication to accurately estimate the channel in the RIS-assisted high-speed rail communication scenario.

[0006] A channel estimation method for intelligent surface-assisted high-speed rail wireless communication includes:

[0007] Step S1, establishing a communication system for intelligent surface-assisted high-speed rail wireless communication, where the communication system includes a mobile relay, an intelligent surface, and a base station. The base station is linearly deployed along the railway track of the high-speed rail, the mobile relay is deployed on the roof of the high-speed rail, and the intelligent surface is deployed on the lamp posts around the base station;

[0008] Step S2, establishing a link channel model based on the communication system, where the link channel model includes a direct channel from the base station to the mobile relay and a cascaded channel reflected from the base station to the intelligent surface and then to the mobile relay;

[0009] Step S3, through the established link channel model, using the least squares algorithm for channel estimation to obtain noisy channel data;

[0010] Step S4, construct a channel estimation network model based on deep learning, introduce the network structure of the Laplacian pyramid into the channel estimation network model, and the network structure includes multiple levels of generative adversarial networks;

[0011] Step S5, take the noisy channel data as the input of the channel estimation network model. In the channel estimation network model, regard the pilot signal and the channel matrix as two-channel images with different sizes, optimize the loss function of the channel estimation network model, calculate the estimation error according to the estimated channel and backpropagate to train the model parameters until the model parameters converge, so as to obtain the trained channel estimation network model;

[0012] Step S6, use the trained channel estimation network model for channel estimation.

[0013] According to the channel estimation method for intelligent reflecting surface assisted high-speed railway wireless communication provided by the present invention, first establish a link channel model based on the communication system. The link channel model includes the direct channel from the base station to the mobile relay and the cascaded channel reflected from the base station to the intelligent reflecting surface and then to the mobile relay. Then, use the least squares algorithm to perform a preliminary estimation of the channel to obtain the noisy channel data. Next, construct a channel estimation network model based on deep learning, take the noisy channel data as the input of the channel estimation network model. In the channel estimation network model, regard the pilot signal and the channel matrix as two-channel images with different sizes. The real part and the imaginary part of the complex matrix can respectively correspond to the two channels of the image. In this way, the channel estimation problem can be transformed into a channel image generation problem. By using a generative adversarial network (GAN) to generate the channel image, through the adversarial training of the generator and the discriminator, the residuals between each layer of the Laplacian pyramid and the adjacent layer can be learned, and the channel image from rough to precise can be reconstructed layer by layer. By optimizing the loss function, the accuracy of channel estimation can be further improved, and at the same time, the correct direction of generator optimization is ensured. The present invention can achieve accurate channel estimation results with a small number of pilots, can significantly improve the communication efficiency, and is more suitable for the scenario of RIS-assisted high-speed railway communication. Description of the Drawings

[0014] Figure 1 It is a flowchart of the channel estimation method for intelligent reflecting surface assisted high-speed railway wireless communication provided by the embodiment of the present invention. Detailed Embodiment

[0015] The embodiments of the present invention will be described in detail below. The examples of the embodiments are shown in the drawings, where the same or similar reference numerals denote the same or similar elements or elements with the same or similar functions from beginning to end. The embodiments described below with reference to the drawings are exemplary and are intended to explain the embodiments of the present invention and should not be construed as a limitation of the present invention.

[0016] Please refer toFigure 1 , an embodiment of the present invention provides a channel estimation method for intelligent metasurface-assisted high-speed rail wireless communication, including steps S1 to S6:

[0017] Step S1, establish a communication system for intelligent metasurface-assisted high-speed rail wireless communication. The communication system includes a mobile relay, an intelligent metasurface, and a base station. The base station is linearly deployed along the railway track of the high-speed rail. The mobile relay is deployed on the roof of the high-speed rail, and the intelligent metasurface is deployed on the lamp posts around the base station.

[0018] Among them, each base station (BS) is equipped with M antennas. To avoid the penetration loss of the train carriage and reduce frequent group handovers, a mobile relay (Mobile Relay, MR) is deployed on the roof, equipped with P antennas, responsible for communicating with the base station. To improve the spectral efficiency, intelligent metasurfaces are symmetrically deployed on the lamp posts around the base station. Each RIS consists of N passive reflection elements, which adjust the propagation of the signal to the target direction by changing the amplitude and phase shift of the incident signal. In this embodiment, both the BS antenna and the MR antenna are modeled as a Uniform Linear Array (ULA) architecture. Since the RIS elements are arranged in a plane, the RIS is modeled as a Uniform Platform Array (UPA) architecture.

[0019] Step S2, establish a link channel model based on the communication system. The link channel model includes the direct channel from the base station to the mobile relay, and the cascaded channel reflected from the base station to the intelligent metasurface and then to the mobile relay.

[0020] Among them, the cascaded channel includes the link channel between the base station and the intelligent metasurface, and the channel of the link between the intelligent metasurface and the mobile relay.

[0021] Since both the BS and the RIS are fixed in specific positions in advance, the link between the base station and the intelligent metasurface can be a Line of Sight (LOS) wireless propagation.

[0022] For the link between the intelligent metasurface and the mobile relay, since the LOS wireless transmission channel between the RIS and the MR may be blocked by trees or buildings in actual situations, the channel of the link between the intelligent metasurface and the mobile relay is modeled as a Rice channel.

[0023] Specifically, the established link channel model satisfies the following conditional expressions:

[0024] ;

[0025] Among them, Represents the combined channel of the downlink channel model in the th time slot, represents the link channel between the base station and the intelligent metasurface, represents the reflection coefficient matrix of the intelligent metasurface, represents the th time slot of the channel between the intelligent metasurface and the mobile relay link, represents the th time slot of the direct channel;

[0026] ;

[0027] = ;

[0028] ;

[0029] ;

[0030] Among them, represents the path complex gain of the link channel between the base station and the intelligent metasurface, represents the receiving array response of the intelligent metasurface, and are the steering vectors of the intelligent metasurface along the x-axis and y-axis respectively, is the Kronecker product, and represent the elevation angle and azimuth angle of the signal arriving at the intelligent metasurface respectively, and are the number of reflection units arranged along the x-axis and y-axis on the three-dimensional Cartesian coordinate plane respectively, and represent the first receiving array response parameter and the second receiving array response parameter respectively, represents the distance between two adjacent intelligent metasurfaces, represents the receiving signal wavelength;

[0031] ;

[0032] ;

[0033] ;

[0034] ;

[0035] Among them, is the Rice factor, is the line-of-sight radio propagation component in the channel between the intelligent metasurface and the mobile relay link in the th time slot, is the non-line-of-sight radio propagation component in the intelligent metasurface and mobile relay link at the th time slot, represents the path complex gain of the channel between the intelligent metasurface and the mobile relay link, is an imaginary number, is the Doppler shift of the channel between the intelligent metasurface and the mobile relay link, is the duration of one time slot, represents the receive array response of the intelligent metasurface and mobile relay link, and represent the steering vectors of the reflection link along the x-axis and y-axis respectively, and represent the elevation angle and azimuth angle at which the signal leaves the intelligent metasurface respectively, and represent the steering vectors of the intelligent metasurface along the x-axis and y-axis respectively, represents the speed of the high-speed train;

[0036] ;

[0037] Among them, is a diagonal matrix, represents the amplitude reflection coefficient vector of the intelligent metasurface, and represent the reflection phase shifts of the first passive reflection element and the th passive reflection element in the intelligent metasurface respectively;

[0038] ;

[0039] ;

[0040] Among them, represents the path complex gain of the direct channel, is the Doppler shift of the direct channel, and represent the elevation angle and azimuth angle when the signal arrives at the mobile relay respectively.

[0041] Step S3, through the established link channel model, the least squares algorithm is used for channel estimation to obtain the noisy channel data.

[0042] Among them, based on the established link channel model, the joint channel of the link channel model at all time slots is obtained, so as to obtain the joint channel matrix , and then the least squares algorithm (LS) is used for channel estimation. The cost function of the least squares algorithm is expressed as:

[0043] ;

[0044] Among them, represents the matrix of received signals, represents the matrix of transmitted signals, represents the transpose.

[0045] The core idea of channel estimation using the LS algorithm is to utilize the pilot symbol sequence, solve through constructing a training matrix and the least squares method, and obtain the estimated channel impulse response value. According to the principle of channel estimation, to obtain high-precision channel state information (CSI), the cost function needs to be minimized. Therefore, by finding the minimum value of the above cost function, taking the partial derivative of the cost function and setting the result to 0, and then solving this, the estimated value of the LS algorithm can be obtained.

[0046] Specifically, solve the cost function to obtain the estimated value of the least squares algorithm , and take as the noisy channel data, The expression of

[0047] .

[0048] Step S4, construct a channel estimation network model based on deep learning, introduce the network structure of the Laplacian pyramid into the channel estimation network model, and the network structure includes multiple different levels of generative adversarial networks.

[0049] Among them, in this embodiment, the pilot signal and the channel matrix are regarded as two-channel images with different sizes, and the time domain and the frequency domain respectively correspond to the two channels of the image. In this way, the channel estimation problem is transformed into a channel image generation problem. Then, the image pyramid that realizes multi-scale expression of the image in the image scaling technology is introduced to assist in channel reconstruction. Through the constructed channel estimation network model based on deep learning, the signal received by the BS is preprocessed by the LS algorithm to achieve a preliminary estimation of the channel; then, the preprocessed signal is regarded as a two-channel image as a conditional input, and the channel estimation problem is transformed into the problem of restoring a high-resolution image with a given low-resolution image. By learning these data, the channel matrix is reconstructed layer by layer through each layer of the pyramid.

[0050] Step S5, use the noisy channel data as the input of the channel estimation network model. In the channel estimation network model, regard the pilot signal and the channel matrix as two-channel images with different sizes, optimize the loss function of the channel estimation network model, calculate the estimation error according to the predicted channel and backpropagate to train the model parameters until the model parameters converge, so as to obtain the trained channel estimation network model.

[0051] Among them, in the channel estimation network model, take As the input of the generator, in the generator, a convolutional layer and a ReLU layer are used to extract the low-level features of the received signal; then, an upsampling module is used to amplify it to the size of the real channel, and multiple denoising blocks are cascaded to achieve denoising. Each denoising block has a structure of Conv2d+BN+ReLU and is filtered by a filter with a size of 64×3×3; finally, a combination of Conv2d+BN+ReLU and Conv2d is used; the residual information generated by the generator is input into the discriminator for judgment. The discriminator is designed as a fully connected neural network and has a powerful non-linear modeling ability. Specifically, the discriminator includes 3 fully connected layers, a BN layer, an activation function layer, a Dropout layer, and a Flatten layer. First, the features are non-linearly transformed through 3 fully connected layers to extract the associations between the features. The BN layer is added to normalize the input of each unit to stabilize the learning and help the gradient propagation at the same time. The LeakyReLU is added as the activation function layer, and the Dropout layer is added to prevent overfitting, improve the training speed of the model, and reduce the network parameters; then, a Flatten layer is used to transform the channel matrix into a one-dimensional vector; finally, a scalar is output through a fully connected layer and a Sigmoid function, indicating that the input residual matrix is real rather than a fake residual channel matrix generated by the generator. The learning purpose of the generator is to make the generated residuals as close as possible. The subsequent pyramid repeats this process. Through multiple iterative trainings of the generator and the discriminator, the network finally reaches the Nash equilibrium state and the network model converges.

[0052] Finally, the channel reconstructed by the channel estimation network model satisfies the following formula:

[0053] ;

[0054] where represents the total number of layers of the Laplacian pyramid, represents the i residual generated by the generator on the

[0055] In this embodiment, the loss function of the channel estimation network model is:

[0056] ;

[0057] ;

[0058] ;

[0059] where represents the loss function of the generative adversarial network, represents the supplementary loss function, Denote the generator, denote the discriminator, denote taking the expected value, denote the i channel generated by the Laplacian pyramid at the denote the i channel generated by the Laplacian pyramid at the

[0060] The above loss function , by introducing a supplementary loss function , can ensure the correct direction of optimizing the generator. By designing the above loss function , the accuracy can be further improved.

[0061] Step S6, perform channel estimation using the trained channel estimation network model.

[0062] Among them, the optimal channel estimation scheme can be determined using the trained channel estimation network model, and finally channel estimation is achieved.

[0063] According to the channel estimation method for intelligent reflecting surface assisted high-speed rail wireless communication in the above embodiments, first establish a link channel model based on the communication system. The link channel model includes the direct channel from the base station to the mobile relay and the cascaded channel reflected from the base station to the intelligent reflecting surface and then to the mobile relay. Then, use the least squares algorithm to perform a preliminary estimation of the channel to obtain noisy channel data. Next, construct a channel estimation network model based on deep learning, and use the noisy channel data as the input of the channel estimation network model. In the channel estimation network model, the pilot signal and the channel matrix are regarded as two-channel images with different sizes, and the real part and the imaginary part of the complex matrix can respectively correspond to the two channels of the image. In this way, the channel estimation problem can be transformed into a channel image generation problem. Generate the channel image by using a generative adversarial network (GAN). Through the adversarial training of the generator and the discriminator, learn the residuals between adjacent layers at each layer of the Laplacian pyramid, and gradually reconstruct the channel image from rough to precise. By optimizing the loss function, the accuracy of channel estimation can be further improved, and at the same time, the correct direction of optimizing the generator is ensured. The present invention can achieve accurate channel estimation results using a small number of pilots, can significantly improve the communication efficiency, and is more suitable for the scenario of RIS-assisted high-speed rail communication.

[0064] The above-described embodiments only represent several implementation manners of the present invention. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the present invention patent should be subject to the appended claims.

Claims

1. A channel estimation method for high-speed rail wireless communication assisted by intelligent metasurface, characterized in that: include: Step S1, establishing a communication system for intelligent metasurface-assisted high-speed rail wireless communication, the communication system comprising a mobile relay, an intelligent metasurface and a base station, the base station being deployed linearly along the rails of the high-speed rail, the mobile relay being deployed on the roof of the high-speed rail, and the intelligent metasurface being deployed on lamp posts around the base station; Step S2, establishing a link channel model based on the communication system, wherein the link channel model includes a direct channel from the base station to the mobile relay, and a cascade channel from the base station to the smart metasurface and then to the mobile relay for reflection; Step S3, using the established link channel model and a least squares algorithm to perform channel estimation to obtain noisy channel data; Step S4, constructing a channel estimation network model based on deep learning, introducing a Laplace pyramid network structure into the channel estimation network model, wherein the network structure includes a plurality of generative adversarial networks at different levels; Step S5, using the noisy channel data as the input of the channel estimation network model, in the channel estimation network model, using the pilot signal and the channel matrix as dual-channel images of different sizes, optimizing the loss function of the channel estimation network model, calculating the estimation error according to the estimated channel and back-propagating the training model parameters until the model parameters converge, thereby obtaining a trained channel estimation network model; Step S6: performing channel estimation using the trained channel estimation network model.

2. The channel estimation method for intelligent metasurface-assisted high-speed rail wireless communication according to claim 1 is characterized in that: The cascade channel includes a link channel between the base station and the smart metasurface, and a channel between the smart metasurface and the mobile relay link. The established link channel model satisfies the following conditional formula: ； in, Indicates The joint channel of the downlink channel model of time slots, represents the link channel between the base station and the smart metasurface, represents the reflection coefficient matrix of the smart metasurface, Indicates The channel between the smart metasurface and the mobile relay link in time slots, Indicates Direct channel under time slots; ； = ； ； ； in, represents the path complex gain of the link channel between the base station and the smart metasurface, represents the receiving array response of the smart metasurface, and The intelligent hypersurface is the guiding vector along the x-axis and y-axis respectively, is the Kronecker product, and Respectively represent the pitch angle and azimuth angle of the signal reaching the smart metasurface, and are the number of reflection units placed along the x-axis and y-axis on the three-dimensional Cartesian coordinate plane, and represent the first receiving array response parameter and the second receiving array response parameter respectively, represents the distance between two adjacent smart metasurfaces, Indicates the wavelength of the received signal; ； ； ； ； in, is the Rice factor, For the The line-of-sight wireless propagation component in the link between the smart metasurface and mobile relay in time slots, For the The non-line-of-sight wireless propagation component in the link between the smart metasurface and the mobile relay in time slots. represents the path complex gain of the channel between the smart metasurface and the mobile relay link, is an imaginary number, The Doppler shift of the channel between the smart metasurface and the mobile relay link, is the duration of a time slot, represents the receiving array response of the smart metasurface and the mobile relay link, and represent the steering vectors of the reflection link along the x-axis and y-axis respectively, and Respectively represent the elevation angle and azimuth angle of the signal leaving the smart metasurface, and represent the first receiving array response parameter and the second receiving array response parameter of the smart metasurface to mobile relay link, respectively, Indicates the speed of the high-speed rail; ； in, is a diagonal matrix, represents the amplitude reflection coefficient vector of the smart metasurface, and Respectively represent the first passive reflective element and the first passive reflective element in the smart metasurface. The reflection phase shift of a passive reflection element; ； ； in, represents the path complex gain of the direct channel, is the Doppler shift of the direct channel, and They respectively represent the elevation angle and azimuth angle when the signal reaches the mobile relay.

3. The channel estimation method for intelligent metasurface-assisted high-speed rail wireless communication according to claim 2 is characterized in that: Based on the established link channel model, the joint channel of the link channel model in all time slots is obtained, thereby obtaining the joint channel matrix , and then the least squares algorithm is used for channel estimation. The cost function of the least squares algorithm is The expression is: ； in, represents the matrix of the received signal, represents the matrix of the transmitted signal, represents transpose; For the cost function Solve and get the estimated value of the least squares algorithm ,Will As noisy channel data, The expression is: 。 4. The channel estimation method for intelligent metasurface-assisted high-speed rail wireless communication according to claim 3 is characterized in that: In the channel estimation network model, As the input of the generator, a convolution layer and a ReLU layer are used to extract the low-level features of the received signal; then, an upsampling module is used to It is enlarged to the size of the real channel, and multiple denoising blocks are connected in series to achieve denoising. Each denoising block has a Conv2d+BN+ReLU structure and is filtered with a filter of size 64×3×3. Finally, a combination of Conv2d+BN+ReLU and Conv2d is used. The residual information generated by the generator is input into the discriminator for judgment. The discriminator includes 3 fully connected layers, a BN layer, an activation function layer, a Dropout layer, and a Flatten layer.

5. The channel estimation method for intelligent metasurface-assisted high-speed rail wireless communication according to claim 4 is characterized in that: Channel estimation network model reconstructs the channel Satisfy the following formula: ； in, represents the total number of layers of the Laplacian pyramid, Indicates i The residuals generated by the generator on the layer Laplacian pyramid.

6. The channel estimation method for intelligent metasurface-assisted high-speed rail wireless communication according to claim 5 is characterized in that: Loss function of the channel estimation network model for: ； ； ； in, represents the loss function of the generative adversarial network, represents the complementary loss function, represents a generator, represents the discriminator, It means taking the expected value. Indicates i The channels generated by the layer Laplacian pyramid, Indicates i -1 channel generated by Laplacian pyramid.

Images