Single-bit IRS multi-user communication system channel estimation method based on improved U-shaped network

Abstract

The invention provides a single-bit IRS multi-user communication system channel estimation method based on an improved U-shaped network, and relates to the technical field of wireless communication. The method comprises the specific steps that firstly, a data set is manufactured through simulation, and a training set and a test set are divided; secondly, designing an improved U-shaped network; secondly, training and testing the improved U-shaped network; and finally, evaluating the performance of the proposed channel estimation method. Experiments prove that the U-shaped network designed and improved by the invention can realize channel estimation of a single-bit intelligent reflective surface (IRS) multi-user communication system.

Description

Technical Field

[0001] The invention relates to the technical field of wireless communications, and in particular to a channel estimation method for a single-bit IRS multi-user communication system based on an improved U-shaped network. Background Art

[0002] In recent years, intelligent reflective surfaces (IRS) have been proposed as one of the key enabling technologies for the fifth-generation network due to their powerful ability to digitally control signal reflection in real time through a large number of passive reflective elements, actively combat wireless channel fading and interference, and shape the wireless channel between users and base stations (BS). The IRS multi-user communication system equipped with a high-resolution analog to digital converter (ADC) has the disadvantages of high hardware complexity and high energy consumption. The IRS multi-user communication system equipped with a single-bit ADC (referred to as a single-bit IRS multi-user communication system) can effectively reduce the hardware complexity and energy consumption.

[0003] Channel estimation is one of the key technologies used to evaluate and compensate for channel characteristics in wireless communication systems. Channel estimation provides a basis for signal equalization and demodulation by accurately estimating channel parameters such as gain, phase, fading, multipath effect and noise, thereby improving system performance. Accurate channel estimation helps reduce bit error rate, improve transmission rate and communication quality. In a single-bit IRS multi-user communication system, channel estimation can effectively adapt to changes in the channel environment, reduce the impact of quantization errors, reduce interference, optimize power control and beamforming, and thus improve spectrum efficiency.

[0004] Neural networks have powerful nonlinear modeling and automatic feature learning capabilities. They can learn implicit signal patterns from complex channel environments (such as high noise and non-ideal conditions) and accurately estimate channel parameters. This makes neural networks have great potential in channel estimation in single-bit IRS multi-user communication systems. Summary of the invention

[0005] The present invention proposes a channel estimation method for a single-bit IRS multi-user communication system based on an improved U-type network, and relates to the technical field of wireless communication. The specific steps include: first, making a data set through simulation, dividing a training set and a test set; then, designing an improved U-type network; secondly, training and testing the improved U-type network; finally, evaluating the performance of the proposed channel estimation method.

[0006] The scheme of the present invention is as follows:

[0007] A: Create a data set through simulation and divide it into training set and test set;

[0008] A1: Create a data set through simulation;

[0009] A2: Divide the training set and test set;

[0010] B: Design of improved U-shaped network;

[0011] B1: Design preprocessing module;

[0012] B2: Design encoder;

[0013] B3: Design decoder;

[0014] B4: Design attention module;

[0015] B5: Design output module;

[0016] C: Training and testing the improved U-shaped network;

[0017] C1: training improved U-type network;

[0018] C2: Test the improved U-shaped network;

[0019] D: Evaluate the performance of the proposed channel estimation method.

[0020] The beneficial effects brought by the method proposed by the present invention include at least:

[0021] (1) A channel estimation method for a single-bit IRS multi-user communication system based on an improved U-type network. The present invention uses channel attention to design an improved U-type network and converts the roughly estimated signal matrix As network input, iterative learning starts from a rough estimate of the channel matrix To the channel matrix H k The mapping finally outputs the accurately estimated channel matrix In the encoder of the improved U-shaped network, a dilated convolution residual module is designed to efficiently utilize information and improve the expressiveness of the model; in the decoder of the improved U-shaped network, a dense residual module is designed to improve the feature reconstruction capability and more accurately restore the channel matrix; at the cross-hop connection of the improved U-shaped network, an attention module is designed to enhance the ability to focus on key channel information and suppress irrelevant noise.

[0022] (2) A channel estimation method for a single-bit IRS multi-user communication system based on an improved U-type network. The U-type network designed and improved by the present invention can realize channel estimation for a single-bit IRS multi-user communication system. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is an overall flow chart of a single-bit IRS multi-user communication system channel estimation method based on an improved U-type network involved in the present invention;

[0024] Figure 2 It is a system scenario model diagram of a single-bit IRS multi-user communication system channel estimation method based on an improved U-shaped network involved in the present invention;

[0025] Figure 3 It is an overall network structure diagram of a single-bit IRS multi-user communication system channel estimation method based on an improved U-shaped network involved in the present invention;

[0026] Figure 4 It is a specific structural diagram of a dilated convolution residual module involved in an improved U-shaped network involved in the present invention;

[0027] Figure 5 It is a specific structural diagram of a dense residual module involved in an improved U-type network involved in the present invention;

[0028] Figure 6 It is a specific structural diagram of an attention module involved in an improved U-shaped network involved in the present invention. DETAILED DESCRIPTION

[0029] To facilitate understanding, the implementation process of the present invention is further described in detail in combination with the technical solution and the accompanying drawings.

[0030] A single-bit IRS multi-user communication system channel estimation method based on an improved U-type network, the flow chart of its main steps is as shown in the attached Figure 1 Specifically, it includes the following steps:

[0031] A: Create a data set through simulation and divide it into training set and test set; it includes two main processes, namely, creating a data set through simulation and dividing it into training set and test set:

[0032] A1: Create a data set through simulation; this includes three main processes: building a channel model, calculating the quantized received signal matrix, and calculating a roughly estimated channel matrix;

[0033] (1) Constructing channel model; channel matrix The calculation formula is

[0034] H k =[d k ,B k ]

[0035] In the formula, k represents the user's serial number, such as Figure 2 As shown, represents the direct channel vector of the kth user-base station link, represents the cascaded channel matrix of the kth user-intelligent reflecting surface (IRS)-base station link, M represents the total number of base station antennas, N represents the total number of reflective elements of the IRS, represents the complex field, B k The calculation formula is

[0036] B k =GF k

[0037] In the formula, represents the channel matrix of the IRS-base station link, represents the channel matrix of the kth user-IRS link, and the calculation formula of G is

[0038]

[0039] In the formula, p G represents the path loss of the IRS-base station link, G LoS and G NLoS G represents the line of sight (LoS) channel matrix and non-line of sight (NLoS) channel matrix of the IRS-base station link, respectively. NLoS The elements of are complex Gaussian random variables, G LoS The calculation formula is

[0040]

[0041] In the formula, represents the uniform linear array steering vector of the base station, represents the uniform planar array steering vector of IRS, with the superscript (·) H represents the conjugate transpose operator, and θ 1 denote the arrival azimuth and elevation angle from IRS to the base station, respectively. and θ 2 represent the deviation azimuth and elevation angle from IRS to the base station, respectively. The calculation formula is

[0042]

[0043] Where m represents the ordinal number of the base station antenna. The calculation formula is

[0044]

[0045] Where, d B represents the spacing between antennas at the base station, and λ represents the wavelength of the carrier;

[0046] The calculation formula is

[0047]

[0048] Where n is the ordinal number of the IRS reflector element, The calculation formula is

[0049]

[0050] Where, d I represents the spacing of IRS elements, mod(·) represents the remainder operator, Represents the floor operator, N v represents the number of columns of the uniform planar array of IRS;

[0051] F k The calculation formula is

[0052] F k =diag(f k )

[0053] In the formula, represents the channel vector of the kth user-IRS link, diag(·) represents the diagonal matrix construction operator, and f k The calculation formula is

[0054]

[0055] In the formula, represents the path loss of the kth user-IRS link, and denote the line-of-sight channel vector and non-line-of-sight channel vector of the kth user-IRS link, respectively. The elements of are complex Gaussian random variables, The calculation formula is

[0056]

[0057] In the formula, and denote the arrival azimuth and elevation angle from the kth user to the IRS, respectively;

[0058] (2) Calculate the quantized received signal matrix; In the front-end stage of signal reception and processing, the received signal is first transmitted through the RF link, then quantized by a single-bit ADC, and finally enters the digital signal processor (DSP) for further processing and demodulation. The calculation formula is

[0059]

[0060] Where sgn(·) represents the sign function, Y k represents the received signal matrix, Y k The calculation formula is

[0061] Y k =[y 1,k ,…,y c,k ,…,y C,k ],(c=1,…,C)

[0062] In the formula, C represents the total number of subframes, c represents the ordinal number of the subframe, represents the signal vector received from the kth user in the cth subframe, y c,k The calculation formula is as follows

[0063]

[0064] In the formula, represents the power of each pilot symbol in the pilot sequence, L represents the length of the pilot sequence, and u k represents the pilot sequence that satisfies the pilot orthogonality, that is, represents the pilot matrix during the cth subframe received at the base station, S c The calculation formula is

[0065]

[0066] In the formula, represents the noise matrix that follows Gaussian distribution, p c The calculation formula is

[0067]

[0068] In the formula, β c ∈[0,1] and φ c,n ∈[0,2π] are the phase shift amplitude and phase shift phase corresponding to the nth reflective element of IRS in the cth subframe;

[0069] (3) Calculate a roughly estimated channel matrix; Calculation The formula is

[0070]

[0071] In the formula, the superscript represents the pseudo-inverse matrix operator, The calculation formula is

[0072]

[0073] In the formula, the superscript (·) -1 represents the inverse matrix operator;

[0074] A2: Divide the training set and the test set; the channel matrix H k and the roughly estimated channel matrix H k The constructed data set D is divided into a training set (Training Set) and a test set (Test Set);

[0075] B: Design an improved U-shaped network; as shown in the attached Figure 3 As shown in the figure, it includes 5 main processes, namely designing the preprocessing module, designing the encoder, designing the decoder, designing the attention module and designing the output module:

[0076] B1: Design a preprocessing module; the preprocessing module includes 1 convolution layer and 1 linear rectification function layer (ReLULayer);

[0077] B2: Design the encoder; the encoder of the improved U-shaped network consists of 4 encoding submodules, each of which includes 1 convolution layer and 1 dilated convolution residual module; as shown in the attached Figure 4 As shown in the figure, the dilated convolution residual module includes 2 dilated convolution layers, 2 batch normalization layers, 2 Gaussian error linear unit layers (GELU Layer) and 1 leap connection;

[0078] B3: Design the decoder; the decoder of the improved U-type network consists of three decoding submodules, each of which includes two dense residual modules and one convolutional layer; as shown in the attached figure Figure 5 As shown, the dense residual module includes 2 convolutional layers, 2 batch normalization layers, 2 linear rectification function layers and 3 cross connections;

[0079] B4: Design the attention module; as shown in the attached Figure 6 As shown in Figure 1, the attention module includes 1 convolution layer, 1 batch normalization layer, 1 global average pooling layer, 1 adaptive one-dimensional convolution layer and 1 sigmoid function layer; the convolution kernel size of the adaptive one-dimensional convolution layer is k = ψ(C), and the calculation formula is

[0080] ψ(C)=|log 2 C+1| odd

[0081] Where C represents the number of feature channels of the input adaptive one-dimensional convolutional layer, |·| represents the absolute value operator, and the subscript (·) oddIt represents the nearest odd operator;

[0082] B5: Design the output module; the output module consists of 1 convolutional layer;

[0083] C: Training and testing the improved U-shaped network; including two main processes, namely training the improved U-shaped network and testing the improved U-shaped network;

[0084] C1: Train the improved U-shaped network; input the data samples in the training set into the improved U-shaped network, and use the Adam optimization method to train the improved U-shaped network until the loss converges, save the optimal network model during the training process, and the training ends. The calculation formula of the loss function is:

[0085]

[0086] In the formula, A represents the number of samples in the training set, and the subscript ||·|| F represents the Frobenius norm operator;

[0087] C2: Testing the improved U-shaped network: Using the samples in the test set to test the improved U-shaped network proposed by the present invention, the roughly estimated channel matrix in the test set is As the input of the optimal network model, the final output is the accurately estimated channel matrix

[0088] D: Evaluate the performance of the proposed channel estimation method; by calculating the channel matrix H k With the accurate estimation of the channel matrix The normalized mean square error (NMSE) between is used to evaluate the channel estimation accuracy of the improved U-type network proposed in the present invention. The calculation formula of NMSE is:

[0089]

[0090] Wherein, log(·) represents a logarithmic operator with base 10, E(·) represents an expected value operator, and the unit of NMSE is decibel (dB).

Claims

1. A channel estimation method for a single-bit IRS multi-user communication system based on an improved U-type network, characterized in that The following steps are involved: A: Create a data set through simulation and divide it into training set and test set; It includes two main processes, namely, making a data set through simulation and dividing the training set and test set: A1: Create a data set through simulation; this includes three main processes: building a channel model, calculating the quantized received signal matrix, and calculating a roughly estimated channel matrix; (1) Constructing channel model; channel matrix The calculation formula is H k =[d k ,B k ] In the formula, k represents the user's serial number, represents the direct channel vector of the kth user-base station link, represents the cascaded channel matrix of the kth user-intelligent reflecting surface (IRS)-base station link, M represents the total number of base station antennas, N represents the total number of reflective elements of the IRS, represents the complex field, B k The calculation formula is B k =GF k In the formula, represents the channel matrix of the IRS-base station link, represents the channel matrix of the kth user-IRS link, and the calculation formula of G is In the formula, p G represents the path loss of the IRS-base station link, G LoS and G NLoS G represents the line-of-sight (LoS) channel matrix and non-line-of-sight (NLoS) channel matrix of the IRS-base station link, respectively. NLoS The elements of are complex Gaussian random variables, G LoS The calculation formula is In the formula, represents the uniform linear array steering vector of the base station, represents the uniform planar array steering vector of IRS, with the superscript (·) H represents the conjugate transpose operator, and θ1 represent the arrival azimuth and elevation angle from IRS to the base station, respectively. and θ2 represent the deviation azimuth and elevation angle from the IRS to the base station, respectively. The calculation formula is Where m represents the ordinal number of the base station antenna. The calculation formula is Where, d B represents the spacing between antennas at the base station, and λ represents the wavelength of the carrier; The calculation formula is Where n is the ordinal number of the IRS reflector element, The calculation formula is Where, d I represents the spacing of IRS elements, mod(·) represents the remainder operator, Represents the floor operator, N v represents the number of columns of the uniform planar array of IRS; F k The calculation formula is F k =diag(f k ) In the formula, represents the channel vector of the kth user-IRS link, diag(·) represents the diagonal matrix construction operator, and f k The calculation formula is In the formula, represents the path loss of the kth user-IRS link, and denote the line-of-sight channel vector and non-line-of-sight channel vector of the kth user-IRS link, respectively. The elements of are complex Gaussian random variables, The calculation formula is In the formula, and denote the arrival azimuth and elevation angle from the kth user to the IRS, respectively; (2) Calculate the quantized received signal matrix; In the front-end stage of signal reception and processing, the received signal is first transmitted through the RF link, then quantized by a single-bit ADC, and finally enters the digital signal processor (DSP) for further processing and demodulation. The calculation formula is Where sgn(·) represents the sign function, Y k represents the received signal matrix, Y k The calculation formula is In the formula, C represents the total number of subframes, c represents the ordinal number of the subframe, represents the signal vector received from the kth user in the cth subframe, y c,k The calculation formula is as follows In the formula, represents the power of each pilot symbol in the pilot sequence, L represents the length of the pilot sequence, and u k represents the pilot sequence that satisfies the pilot orthogonality, that is, represents the pilot matrix during the cth subframe received at the base station, S c The calculation formula is In the formula, represents the noise matrix that follows Gaussian distribution, p c The calculation formula is p c =[1,β c exp(jφ c,1 ),…,β c exp(jφ c,n ),…,β c exp(jφ c,N )] In the formula, β c ∈[0,1] and φ c,n ∈[0,2π] are the phase shift amplitude and phase shift phase corresponding to the nth reflective element of IRS in the cth subframe; (3) Calculate a roughly estimated channel matrix; The calculation formula is In the formula, the superscript represents the pseudo-inverse matrix operator, The calculation formula is In the formula, the superscript (·) -1 represents the inverse matrix operator; A2: Divide the training set and the test set; the channel matrix H k and a rough estimate of the channel matrix The constructed data set D is divided into a training set (Training Set) and a test set (Test Set); B: Design of improved U-shaped network; It includes 5 main processes, namely designing the preprocessing module, designing the encoder, designing the decoder, designing the attention module and designing the output module: B1: Design preprocessing module; The preprocessing module includes 1 convolution layer and 1 linear rectification function layer (ReLU Layer); B2: Design the encoder; the encoder of the improved U-shaped network consists of 4 encoding submodules, each of which includes 1 convolution layer and 1 dilated convolution residual module; the dilated convolution residual module includes 2 dilated convolution layers, 2 batch normalization layers, 2 Gaussian error linear unit layers (GELU Layer) and 1 leap connection; B3: Design the decoder; the decoder of the improved U-network consists of 3 decoding submodules, each of which includes 2 dense residual modules and 1 convolutional layer; the dense residual module includes 2 convolutional layers, 2 batch normalization layers, 2 linear rectification function layers and 3 cross connections; B4: Design attention module; The attention module includes 1 convolution layer, 1 batch normalization layer, 1 global average pooling layer, 1 adaptive one-dimensional convolution layer and 1 sigmoid function layer; the convolution kernel size of the adaptive one-dimensional convolution layer is k = ψ(C), and the calculation formula is ψ(C) = |log2 C+1| odd Where C represents the number of feature channels of the input adaptive one-dimensional convolutional layer, |·| represents the absolute value operator, and the subscript (·) odd It represents the nearest odd operator; B5: Design the output module; the output module consists of 1 convolutional layer; C: Training and testing the improved U-shaped network; including two main processes, namely training the improved U-shaped network and testing the improved U-shaped network; C1: Train the improved U-shaped network; input the data samples in the training set into the improved U-shaped network, and use the Adam optimization method to train the improved U-shaped network until the loss converges, save the optimal network model during the training process, and the training ends. The calculation formula of the loss function is In the formula, A represents the number of samples in the training set, and the subscript ||·|| F represents the Frobenius norm operator; C2: Testing the improved U-shaped network: Using the samples in the test set to test the improved U-shaped network proposed by the present invention, the roughly estimated channel matrix in the test set is As the input of the optimal network model, the final output is the accurately estimated channel matrix D: Evaluate the performance of the proposed channel estimation method; by calculating the channel matrix H k With the accurate estimation of the channel matrix The normalized mean square error (NMSE) between is used to evaluate the channel estimation accuracy of the improved U-type network proposed in the present invention. The calculation formula of NMSE is: Wherein, log(·) represents a logarithmic operator with base 10, E(·) represents an expected value operator, and the unit of NMSE is decibel (dB).

Images