Channel estimation method and system for intelligent metasurface auxiliary communication and sensing system

Abstract

The invention belongs to the field of wireless communication, and provides a channel estimation method and system for an intelligent metasurface auxiliary communication and sensing system, and the method comprises the steps: obtaining an original received signal, and carrying out the preprocessing of the original received signal, and obtaining a user side received signal and a base station received signal; obtaining a user side channel estimation result and a base station channel estimation result by using a pre-trained channel estimation network model based on the user side receiving signal and the base station receiving signal; wherein the channel estimation network model comprises a pre-estimation network and a denoising network; the pre-estimation network comprises three layers of feature extraction modules and a full connection layer, and channel attention modules are respectively connected behind the second layer of feature extraction module and the third layer of feature extraction module; the de-noising network comprises a convolutional layer, an activation function layer, a three-layer channel de-noising module, a deep convolutional layer, a jump connection layer and a full connection layer, wherein the jump connection layer is connected with the output of the deep convolutional layer and the input of the de-noising module. And a high-precision channel estimation result can be obtained under relatively low complexity.

Description

Technical Field

[0001] The present invention belongs to the field of wireless communication technology, and specifically relates to a channel estimation method and system for an intelligent metasurface-assisted communication and perception system. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] With the rapid development of wireless communication technology, 6G communication systems are placing higher demands on spectrum efficiency, coverage, and system performance. To address these challenges, RIS, as a new technology, has garnered widespread attention in academia. RIS dynamically adjusts the reflection and refraction properties of electromagnetic waves, thereby enhancing the coverage and communication capacity of wireless communication systems. In particular, in millimeter-wave (mmWave) communications, RIS can effectively compensate for the poor coverage caused by short wavelengths. Furthermore, ISAC technology integrates communication and perception systems to improve spectrum efficiency and system performance. However, obtaining accurate channel state information (CSI) is crucial for achieving efficient and reliable communication in RIS-ISAC systems, and this presents a significant challenge for channel estimation technology.

[0004] Currently, the most commonly used channel estimation algorithm in communication systems is the least squares (LS) method. However, LS suffers from poor noise immunity and, in the presence of high noise levels, results in low estimation accuracy. To address this low accuracy, a two-phase transmission scheme has been proposed. In the first phase, a scanning-based RIS reflection coefficient is used, and in the second phase, optimized RIS reflection coefficients are used to obtain coarse and fine sensing channel estimates, respectively. This improves channel estimation accuracy, but incurs significant training overhead. Another channel estimation method based on the extreme learning machine (ELM) sequentially estimates the direct-line channel and the RIS reflection channel. Compared to methods based on least squares, this method offers higher accuracy and lower model complexity. Another approach models CSI estimation as an image super-resolution problem, using a super-resolution convolutional neural network and a denoising convolutional neural network to recover and denoise the channel matrix, effectively improving estimation accuracy in RIS systems. A conditional generative adversarial network (CGAN) is used for downlink channel estimation, significantly improving accuracy compared to methods based on the ELM. In the RIS-ISAC system, although the above methods improve the accuracy of channel estimation to varying degrees, they all lead to an increase in the complexity of the overall solution. How to obtain high-precision channel estimation results with lower complexity remains one of the problems that need to be solved. Summary of the Invention

[0005] In order to solve the above problems, the present invention proposes a channel estimation method and system for an intelligent metasurface-assisted communication and perception system, which can obtain high-precision channel estimation results with low complexity.

[0006] According to some embodiments, a first solution of the present invention provides a channel estimation method for an intelligent metasurface-assisted communication and perception system, which adopts the following technical solutions: A channel estimation method for an intelligent metasurface-assisted communication and perception system, comprising: Obtain the original received signal and perform preprocessing to obtain the user end received signal and the base station received signal; Based on the user-side received signal and the base station received signal, the user-side channel estimation result and the base station channel estimation result are obtained using the pre-trained channel estimation network model; Wherein, the channel estimation network model includes a pre-estimation network and a denoising network; The pre-estimation network consists of three layers of feature extraction modules and one fully connected layer. A channel attention module is connected to the second and third feature extraction modules respectively. The denoising network includes a convolutional layer, an activation function layer, a three-layer channel denoising module, a depth convolutional layer, a skip connection layer and a fully connected layer. The skip connection layer connects the output of the depth convolutional layer and the input of the denoising module.

[0007] Furthermore, a reconfigurable intelligent metasurface-assisted communication and perception system is used to obtain the original received signal, wherein the system includes a base station with M antennas, a RIS with N reflection units, and K single-antenna users; A first communication link is formed between the base station and the RIS, which is a base station-RIS channel; A second communication link is formed between the RIS and the K single-antenna users, which is the RIS-user channel; The sensing link between the base station and the sensing target and then back to the base station is the base station-sensing target-base station channel. The base station-RIS channel and the RIS-user channel form a reflection cascade channel.

[0008] Furthermore, the channel matrix of the base station-RIS channel , specifically:

[0009] in, is the number of paths between the base station and the RIS, For the The complex gain of each path, and Respectively The arrival and departure angles of the paths, is the conjugate transpose, and are the steering vectors for the arrival and departure angles, respectively.

[0010] Furthermore, the channel matrix from RIS to the kth user is , specifically:

[0011] in, For RIS and the kth user The number of paths between and Respectively The complex gain and arrival angle of each path, is the steering vector.

[0012] Furthermore, the obtaining of the original received signal and preprocessing thereof is specifically as follows: Acquire original received signals, which include user original received signals and base station original received signals; Separate the real and imaginary parts of the user's original received signal into two channels to obtain the user-side received signal; The real part and imaginary part of the original base station received signal are separated into two channels to obtain the base station received signal.

[0013] Furthermore, the feature extraction module includes a convolution layer, a normalization layer and an activation function ReLU; The channel attention module includes a maximum average pooling layer, a multi-layer perceptron layer, and an activation function layer. The channel attention weight matrix output by the activation function layer is multiplied by the input features of the channel attention module to obtain the output of the channel attention module.

[0014] Furthermore, the maximum average pooling layer includes a parallel global maximum pooling layer and a global average pooling layer; The outputs of the global maximum pooling layer and the global average pooling layer are processed by the multi-layer perceptron layer and then concatenated as the input of the activation function layer.

[0015] Furthermore, the channel denoising module includes a first denoising path, a second denoising path, a connection layer, a convolutional layer, a channel attention module and a skip connection; The first denoising path and the second denoising path both include two convolution blocks with different dilation rates, and an activation function is connected after the two convolution blocks.

[0016] Furthermore, the first denoising path and the second denoising path respectively process input characteristics of the channel denoising module; The output of the first denoising path and the output of the second denoising path are concatenated through a connection layer.

[0017] According to some embodiments, a second solution of the present invention provides a channel estimation system for an intelligent metasurface-assisted communication and perception system, which adopts the following technical solutions: The channel estimation system of the intelligent metasurface-assisted communication and perception system includes: The signal preprocessing module is configured to obtain the original received signal and perform preprocessing to obtain the user end received signal and the base station received signal; The channel estimation module is configured to obtain a user-side channel estimation result and a base station channel estimation result based on the user-side received signal and the base station received signal using a pre-trained channel estimation network model; Wherein, the channel estimation network model includes a pre-estimation network and a denoising network; The pre-estimation network consists of three layers of feature extraction modules and one fully connected layer. A channel attention module is connected to the second and third feature extraction modules respectively. The denoising network includes a convolutional layer, an activation function layer, a three-layer channel denoising module, a depth convolutional layer, a skip connection layer and a fully connected layer. The skip connection layer connects the output of the depth convolutional layer and the input of the denoising module.

[0018] Compared with the prior art, the present invention has the following beneficial effects: The present invention first simplifies the received signal using orthogonal pilots, then performs channel estimation using a pre-estimation module and a denoising module. In the denoising module, features are extracted from different receptive fields and weighted using a channel attention mechanism. The user-side received signal and the base station-side echo signal are then input into the network to produce trained networks. The trained networks are then tested to obtain the final channel estimation results. The test results show that the network can simultaneously estimate the communication channel and the cascade channel, with high accuracy for both the communication cascade channel and the perception channel, and maintains good performance even when the number of base station and user antennas changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0020] Figure 1 2. This is a diagram of a downlink communication system of a reconfigurable intelligent metasurface-assisted communication and perception system according to an embodiment of the present invention; Figure 2 It is the transmission protocol of the reconfigurable intelligent metasurface assisted communication and perception system in the embodiment of the present invention; Figure 3 This is a diagram showing the main architecture of a channel estimation network model according to an embodiment of the present invention; Figure 4 is a channel attention module in an embodiment of the present invention; Figure 5 This is the channel denoising module structure in an embodiment of the present invention; Figure 6 It is an estimation scheme of the channel estimation network model in an embodiment of the present invention; FIG7( a ) is a comparison of NMSE cascade channels of different channel estimation schemes according to an embodiment of the present invention; FIG7( b ) is a comparison of NMSE-perceived channels of different channel estimation schemes according to an embodiment of the present invention; FIG8( a ) is a comparison of NMSE cascade channels with different numbers of base station antennas according to an embodiment of the present invention; FIG8( b ) is a comparison of NMSE-sensed channels for different numbers of base station antennas in an embodiment of the present invention. DETAILED DESCRIPTION

[0021] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0022] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0023] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0024] In the absence of conflict, the embodiments of the present invention and the features thereof may be combined with each other.

[0025] Example 1 This embodiment provides a channel estimation method for an intelligent metasurface-assisted communication and perception system. In this embodiment, the method includes the following steps: Obtain the original received signal and perform preprocessing to obtain the user end received signal and the base station received signal; Based on the user-side received signal and the base station received signal, the user-side channel estimation result and the base station channel estimation result are obtained using the pre-trained channel estimation network model; Wherein, the channel estimation network model includes a pre-estimation network and a denoising network; The pre-estimation network consists of three layers of feature extraction modules and one fully connected layer. A channel attention module is connected to the second and third feature extraction modules respectively. The denoising network includes a convolutional layer, an activation function layer, a three-layer channel denoising module, a depth convolutional layer, a skip connection layer and a fully connected layer. The skip connection layer connects the output of the depth convolutional layer and the input of the denoising module.

[0026] The training process of the method described in this embodiment is as follows: Step 1: Model the reconfigurable intelligent metasurface-assisted communication and perception system, including base stations, users, RIS, and perception targets.

[0027] like Figure 1As shown, the RIS-ISAC system is modeled, and the Saleh-Valenzuela channel model is used to establish a downlink RIS-Wave multiple input multiple output (MIMO) system model. The base station (BS) sends orthogonal pilots through the RIS to the user equipment (UE). For the sensing link, the base station sends orthogonal pilots to the sensing target and receives the echo signal. Therefore, the sensing link can be modeled as a MIMO system in which a multi-antenna base station sends signals and a multi-antenna base station receives signals. The BS of the ISAC system can process communication and sensing link information. The figure contains three channels: the BS-RIS channel, the RIS-UE channel, and the BS-target-BS channel. The number of base station antennas is M, the number of RIS reflection units is N, and the system contains K single-antenna users. Specifically: The system consists of a fixed-position multi-antenna BS, a fixed-position RIS, multiple single-antenna UEs, and a sensing target; There are three channels in the system: BS-target-BS, BS-RIS, and RIS-UE. The BS-RIS channel and the RIS-UE channel are combined into a reflection cascade channel. The line-of-sight channel between the BS and the UE is blocked by obstacles and cannot directly transmit signals; The base station sends a pilot signal along one path to the target and receives an echo signal. The other path is reflected by the RIS and reaches the UE. For each path, the received signal is obtained based on the channel matrix.

[0028] Step 1-1: Both the base station and the RIS reflector unit use a uniform linear array (ULA). First, establish the channel model and the channel matrix between the BS and RIS. As shown in formula (1): (1) in, is the number of paths between the base station and the RIS, For the The complex gain of each path, and Respectively The angle of arrival and angle of departure of each path. is the conjugate transpose, and are the steering vectors for the arrival and departure angles, respectively.

[0029] The channel model from RIS to the kth user is: As shown in formula (2): (2) in, For RIS and The number of paths between . and Respectively The complex gain and arrival angle of each path. is the steering vector, as shown in formula (3): (3) in, is the number of antennas or elements, is transposed. for The direction cosines corresponding to the angle are shown in formula (4): (4) in, is the carrier wavelength, is the component spacing, assuming The channel from the base station to the sensing target and then to the base station As shown in formula (5): (5) in is the reflection coefficient.

[0030] Step 1-2: This embodiment can be extended from UPL to uniform planar array (UPA). The UPA, which is the total number of base station antennas or RIS units, and The UPA array takes the elevation angle into consideration. and azimuth , whose array steering vector As shown in formula (6): (6) in, is the Kronecker product, and The expressions of are shown in formula (7) and formula (8): (7) (8) in, and The expressions of are shown in formula (9) and formula (10): (9) (10) Step 1-3: If Figure 1 As shown, when the direct channel between the BS and the UE is blocked by an obstacle, the signal cannot be transmitted directly. The signal can be reflected by the RIS. The RIS can adjust the amplitude and phase of the incident signal. The reflection matrix of the RIS is As shown in formula (11): (11) in, and are the amplitude and phase shift of the nth passive reflection unit in RIS, respectively.

[0031] Step 1-4: If Figure 2 As shown, consider C subframes, each of which includes a pilot for channel estimation and data to be transmitted. ) frame to send orthogonal pilots of length B. Set the orthogonal pilots to , the pilot in the cth frame As shown in formula (12): (12) Because the pilot sequence is the same for different frames, the RIS phase shift remains unchanged in the same frame. The RIS phase shift in the cth frame is , the total RIS phase shift within the C frame As shown in formula (13): (13) After the base station transmits the signal, the user U at the user end k The received signal in the pth time slot , as shown in formula (14): (14) in, is the noise power Additive White Gaussian Noise (AWGN) because , so the original formula (14) can be converted to formula (15): (15) in, is the cascade channel GH, so It can be converted into formula (16): (16) Step 1-5: For orthogonal pilot signals exist make .in is the transmission power. Multiply the right side of equation (14) by To eliminate the pilot sequence, the deformed received signal As shown in formula (17): (17) Noise after deformation As shown in formula (18): (18) The transformed formula of formula (14) is shown in formula (19): (19) The received signals in B time slots are superimposed to obtain As shown in formula (20): (20) in, is the reflection coefficient matrix, the noise matrix As shown in formula (21): (twenty one) Step 1-6: For the sensing channel, the base station receives the echo signal in the cth frame As shown in formula (22): (twenty two) in, is the self-interference matrix. Since the propagation environment between the BS transmit antenna and the receive antenna is stable, the self-interference channel can be pre-determined on the BS .eliminate Then, multiply the right side of equation (22) by To eliminate the pilot sequence, the deformed received signal As shown in formula (23): (twenty three) Noise after deformation As shown in formula (24): (twenty four) The transformed formula of formula (22) is shown in formula (25): (25) Among them, the received signals in B time slots are superimposed to obtain , as shown in formula (26): (26) in, is the received signal superimposed in B time slots, is the noise matrix.

[0032] Step 2: Process the original received signal and preprocess the original received signal to facilitate input into the channel estimation network.

[0033] Step 2-1: Take the received signal in the C frame as input and separate the real part and the imaginary part into two channels.

[0034] For the user end receiving the signal, the input data As shown in formula (27): (27) For the base station to receive the signal, the input data As shown in formula (28): (28) Step 2-2: Set the specific parameters as shown in Table 1.

[0035] Table 1 Parameter settings

[0036] Step 3: Design the channel estimation network. The specific network block diagram is as follows Figure 3 As shown, it includes a pre-estimation network and a denoising network.

[0037] Step 3-1: Build a pre-estimation network for preliminary feature extraction and dimension transformation of the received signal. First, estimate the cascade channel and convert the user end received signal Receive signals from base stations After feature extraction, a channel attention (CA) module is built to assign weights to the features. The assigned feature outputs are flattened and input into the fully connected (FC) layer for dimensionality transformation, transforming the dimensions to the cascade channel matrix.

[0038] First, the communication signal and the perception signal are input into the pre-estimation network. The pre-estimation network consists of three feature extraction layers. The first feature extraction layer performs preliminary feature extraction. The feature extraction layer includes a convolution layer, a normalization layer, and an activation function ReLU. The second and third layers are added with a channel attention module after feature extraction. The channel attention module can weight the extracted features so that the network can pay more attention to important features. The channel attention module structure is as follows: Figure 4 shown.

[0039] In the channel attention module, the input first passes through the maximum average pooling layer, which extracts global information for each channel from the input feature map. It is then processed by the multi-layer perceptron (MLP). This layer uses nonlinear transformations to learn the complex relationships between channels and adaptively identify which channels are more important for the task. The sigmoid layer then normalizes the MLP output to the range [0, 1] to generate channel attention weights. After obtaining the channel attention weight matrix, the weighted output features of the channel attention module are obtained by multiplying the channel attention weight matrix with the input feature matrix of the channel attention module.

[0040] The output features of the last layer of channel attention module are processed by the fully connected layer to obtain the pre-estimated channel matrix, including the pre-estimated cascade channel matrix And the pre-estimated sensing channel matrix .

[0041] Step 3-2: Input the pre-estimated channel matrix into the denoising network. The denoising network first extracts features from the input pre-estimated channel matrix and then performs denoising through three series-connected denoising modules (CDM).

[0042] A denoising network is built to deeply denoise the output of the pre-estimation network. First, deep feature extraction is performed through convolutional layers and activation function layers. A lightweight channel denoising module (CDM) is then designed to perform multi-scale feature processing. Each CDM module is divided into two branches at different scales, which are then fused through a concat layer and weighted by a call aggregation module. Three CDMs are connected in series to perform progressive denoising, and skip connections are used to mitigate the vanishing gradient problem. The network output is the cascaded channel estimation result.

[0043] CDM structure is as follows Figure 5 As shown, it includes two paths. In the first path, the convolution blocks with expansion rates of 1 and 2 are set respectively. The input With output The relationship is shown in formula (29): (29) in, is the output of the first convolution layer, k is the convolution kernel size, d is the dilation rate, p is the padding, and s is the stride. On the second path, convolution blocks with dilation rates of 3 and 4 are set respectively. Input With output The relationship is shown in formula (30): (30) in, The output of the first convolutional layer is concatenated through the Concat layer. As shown in formula (31): (31) After getting the output, the concatenated features are weighted and the feature maps are input into the channel attention module. The pooling layer of the channel attention module undergoes maximum pooling and average pooling operations. The output of the maximum pooling is and the average pooling output As shown in formula (32) and formula (33) respectively: (32) (33) The output results are concatenated and input into a shared MLP, which generates unnormalized attention weights after processing. As shown in formula (34): (34) in, is the weight matrix of the first fully connected layer, is the weight matrix of the second fully connected layer, and r is the dimensionality reduction ratio. Then it is normalized by the sigmoid function to generate the final weight As shown in formula (35): (35) By adding the feature And the obtained feature weight matrix Multiply to get the final output As shown in formula (36): (36) Step 3-3: After the output of the denoising module, pass it through a convolution block, then a skip connection to preserve the pre-estimated information, and finally a fully connected layer to convert the dimension to the required channel matrix dimension.

[0044] Step 4: Test the trained channel estimation network. The specific steps are as follows: Figure 6 As shown. Through the input signal and And the built MSDNet is trained. The base station receives the signal And the user end receives the signal Input into the adjusted channel estimation network for pre-estimation and denoising.

[0045] The specific training parameters are shown in Table 2.

[0046] Table 2 Parameter settings for the entire network training

[0047] After the channel estimation network obtains the output, back propagation is performed through the loss function, and the loss function loss is shown in formula (37): (37) in, For the sake of expectation, and are the channel estimation value and the true value respectively.

[0048] Step 5: Test the channel estimation network obtained after training and use the test data and Input them into G-MSDNet and S-MSDNet respectively to obtain the final channel estimation results, that is, and .

[0049] Specifically, the G-MSDNet and S-MSDNet models are the results of training on two different received signals respectively. The model architectures of the two are exactly the same, but the training data are different.

[0050] G-MSDNet is a network obtained by inputting the user-side received signal into the network for training. After training, it is used to estimate the cascade channel and obtain the user-side channel estimation result. .

[0051] S-MSDNet is a network obtained by inputting the base station receiving signal into the network for training. After training, it is used to estimate the sensing channel and obtain the base station channel estimation result. .

[0052] The solution described in this embodiment has a good estimation effect on the communication channel and the perception channel in the RIS-ISAC system. In order to prove the effectiveness of this embodiment, this solution compares the baseline LS, the CNN and DNN networks that can be used for channel estimation, the denoising networks DnCNN and RIDNet with better effects, and the CDM network, where the CDM network is MSDNet with the pre-estimation module removed. Figure 7 shows the NMSE comparison results of the present invention and the other six solutions. It can be seen from Figure 7(a) that the present invention has higher accuracy in estimating cascade channels than other networks. It can be seen from Figure 7(b) that the present invention has a more obvious estimation effect on the perception channel.

[0053] Furthermore, to explore the impact of different numbers of base station antennas and RIS units on network performance, tests were conducted by varying the values ​​of M and N. Figure 8 compares the NMSE performance for different numbers of base station antennas, testing NMSE performance for M values ​​of 8, 12, 16, 20, and 24. Figure 8(a) shows that for cascaded channels, at low signal-to-noise ratios (SNRs), NMSE decreases slightly as M increases. At high SNRs, NMSE increases slightly, but the change is not significant. Figure 8(b) shows that NMSE increases slightly with M.

[0054] Example 2 This embodiment provides a channel estimation system for an intelligent metasurface-assisted communication and perception system, including: The signal preprocessing module is configured to obtain the original received signal and perform preprocessing to obtain the user end received signal and the base station received signal; The channel estimation module is configured to obtain a user-side channel estimation result and a base station channel estimation result based on the user-side received signal and the base station received signal using a pre-trained channel estimation network model; Wherein, the channel estimation network model includes a pre-estimation network and a denoising network; The pre-estimation network consists of three layers of feature extraction modules and one fully connected layer. A channel attention module is connected to the second and third feature extraction modules respectively. The denoising network includes a convolutional layer, an activation function layer, a three-layer channel denoising module, a depth convolutional layer, a skip connection layer and a fully connected layer. The skip connection layer connects the output of the depth convolutional layer and the input of the denoising module.

[0055] The examples and application scenarios implemented by the above modules and corresponding steps are the same, but are not limited to the contents disclosed in the above embodiment 1. It should be noted that the above modules as part of the system can be executed in a computer system such as a set of computer executable instructions.

[0056] The descriptions of the various embodiments in the above embodiments have different focuses. For parts not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0057] The proposed system can be implemented in other ways. For example, the system embodiment described above is merely illustrative. For example, the above module division is only a logical function division. In actual implementation, other division methods may be used. For example, multiple modules can be combined or integrated into another system, or some features can be ignored or not implemented.

[0058] Although the above describes the specific embodiments of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without any creative work are still within the scope of protection of the present invention.

Claims

1. A channel estimation method for an intelligent metasurface-assisted communication and perception system, characterized in that: include: Obtain the original received signal and perform preprocessing to obtain the user end received signal and the base station received signal; Based on the user-side received signal and the base station received signal, the user-side channel estimation result and the base station channel estimation result are obtained using the pre-trained channel estimation network model; Wherein, the channel estimation network model includes a pre-estimation network and a denoising network; The pre-estimation network consists of three layers of feature extraction modules and one fully connected layer. A channel attention module is connected to the second and third feature extraction modules respectively. The denoising network includes a convolutional layer, an activation function layer, a three-layer channel denoising module, a depth convolutional layer, a skip connection layer and a fully connected layer. The skip connection layer connects the output of the depth convolutional layer and the input of the denoising module.

2. The channel estimation method of the intelligent metasurface assisted communication and perception system according to claim 1, characterized in that A reconfigurable intelligent metasurface-assisted communication and perception system is used to obtain the original received signal. The system includes a base station with M antennas, a RIS with N reflection units, and K single-antenna users. A first communication link is formed between the base station and the RIS, which is a base station-RIS channel; A second communication link is formed between the RIS and the K single-antenna users, which is the RIS-user channel; The sensing link between the base station and the sensing target and then back to the base station is the base station-sensing target-base station channel. The base station-RIS channel and the RIS-user channel form a reflection cascade channel.

3. The channel estimation method of the intelligent metasurface assisted communication and perception system according to claim 2, characterized in that: Channel matrix of base station-RIS channel , specifically: in, is the number of paths between the base station and the RIS, For the The complex gain of each path, and Respectively The arrival and departure angles of the paths, is the conjugate transpose, and are the steering vectors for the arrival and departure angles, respectively.

4. The channel estimation method of the intelligent metasurface assisted communication and perception system according to claim 2, characterized in that: Channel matrix from RIS to the kth user , specifically: in, For RIS and the kth user The number of paths between and Respectively The complex gain and arrival angle of each path, is the steering vector.

5. The channel estimation method of the intelligent metasurface assisted communication and perception system according to claim 1, characterized in that: The obtaining of the original received signal and preprocessing thereof are specifically as follows: Acquire original received signals, which include user original received signals and base station original received signals; Separate the real and imaginary parts of the user's original received signal into two channels to obtain the user-side received signal; The real part and imaginary part of the original base station received signal are separated into two channels to obtain the base station received signal.

6. The channel estimation method of the intelligent metasurface assisted communication and perception system according to claim 1, characterized in that: The feature extraction module includes a convolution layer, a normalization layer and an activation function ReLU; The channel attention module includes a maximum average pooling layer, a multi-layer perceptron layer, and an activation function layer. The channel attention weight matrix output by the activation function layer is multiplied by the input features of the channel attention module to obtain the output of the channel attention module.

7. The channel estimation method of the intelligent metasurface assisted communication and perception system according to claim 6, characterized in that: The maximum average pooling layer contains a parallel global maximum pooling layer and a global average pooling layer; The outputs of the global maximum pooling layer and the global average pooling layer are processed by the multi-layer perceptron layer and then concatenated as the input of the activation function layer.

8. The channel estimation method for the intelligent metasurface-assisted communication and perception system according to claim 1, wherein: The channel denoising module includes a first denoising path, a second denoising path, a connection layer, a convolutional layer, a channel attention module and a skip connection; The first denoising path and the second denoising path both include two convolution blocks with different dilation rates, and an activation function is connected after the two convolution blocks.

9. The channel estimation method of the intelligent metasurface assisted communication and perception system according to claim 8, characterized in that: The first denoising path and the second denoising path respectively process input characteristics of the channel denoising module; The output of the first denoising path and the output of the second denoising path are concatenated through a connection layer.

10. Channel estimation system for intelligent metasurface-assisted communication and perception system, characterized in that: include: The signal preprocessing module is configured to obtain the original received signal and perform preprocessing to obtain the user end received signal and the base station received signal; The channel estimation module is configured to obtain a user-side channel estimation result and a base station channel estimation result based on the user-side received signal and the base station received signal using a pre-trained channel estimation network model; Wherein, the channel estimation network model includes a pre-estimation network and a denoising network; The pre-estimation network consists of three layers of feature extraction modules and one fully connected layer. A channel attention module is connected to the second and third feature extraction modules respectively. The denoising network includes a convolutional layer, an activation function layer, a three-layer channel denoising module, a depth convolutional layer, a skip connection layer and a fully connected layer. The skip connection layer connects the output of the depth convolutional layer and the input of the denoising module.

Images