A multi-RIS indoor positioning method based on deep denoising

Abstract

This invention provides a deep denoising-based multi-RIS indoor positioning method. For indoor scenarios with multipath interference and noise, this invention first constructs a fourth-order tensor model from the channel state information reflected from multiple indoor RIS points and scattering points to the receiver. Second, it uses antenna rearrangement technology to rearrange the tensor model to satisfy the decomposition uniqueness condition. Then, it utilizes the denoising capabilities of a deep learning architecture to suppress noise in the received signal. Further, it employs an optimized quadlinear alternating least squares algorithm to estimate the factor matrix and extract channel parameters. Finally, it uses a geometrically based search-free spatial positioning method to estimate the user's location and azimuth angle, and map the indoor environment. Therefore, the deep denoising-based multi-RIS indoor positioning method proposed in this invention has higher accuracy and robustness compared to comparative algorithms, and is more in line with the needs of practical communication scenarios.

Description

Technical Field

[0001] This invention relates to the field of wireless communication technology, and in particular to a multi-RIS indoor positioning method based on deep denoising. Background Technology

[0002] In recent years, deep learning technology has rapidly developed in the field of communication sensing, demonstrating outstanding performance in tasks such as channel estimation, symbol detection, beamforming, and localization. In denoising tasks, deep learning methods can learn the relationship between contaminated observation data and clean signals through end-to-end nonlinear mapping, exhibiting strong robustness. Against this backdrop, deep learning-based denoising techniques are also being considered for denoising tasks in the field of communication sensing, thereby achieving high-precision and robust target perception.

[0003] The presence of complex multipath effects and frequent line-of-sight path obstruction makes indoor positioning extremely challenging. Reconfigurable smart surfaces (RIS) have been introduced to reshape the wireless environment due to their low power consumption and hardware cost advantages. RIS can construct a programmable wireless propagation environment through intelligent control of electromagnetic reflections, thereby improving signal transmission quality and positioning accuracy. Currently, most research focuses on single-RIS scenarios, but in large-scale or heavily obstructed environments, the capabilities of a single RIS may be insufficient to achieve comprehensive coverage. The deployment of multi-RIS assisted systems not only provides richer spatial paths and propagation diversity but also facilitates stable positioning under non-line-of-sight conditions.

[0004] In the paper by K. Li, M. El-Hajjar and M. Xiao (Reconfigurable IntelligentSurface Aided Position and Orientation Estimation Based on Joint Beamforming with Limited Feedback[J]. IEEE Open J. Commun. Soc., vol. 4, pp. 748-767, 2023.), a RIS-assisted millimeter-wave position and orientation estimation scheme based on joint beamforming with limited feedback is adopted. However, the research is limited to the study of a single RIS, which may not provide sufficient coverage under severely obstructed conditions. In the paper by J. Ding, Y. Wang, H. Si, S. Gao and J. Xing (TDLoc: Passive Localization for MIMO-OFDM System via Tensor Decomposition[J]. IEEE Internet Things J., vol.9, no. 21, pp. 21687-21701, Nov. 2022.), a passive indoor 3D moving target localization and tracking method based on mobile hotspot (WIFI) channel state information sensing and a recurrent neural network model was adopted. However, this paper has not yet applied the computing power based on deep learning to signal denoising. Summary of the Invention

[0005] Purpose of the invention: To address the shortcomings of existing technologies, this invention proposes a multi-RIS indoor positioning method based on depth denoising.

[0006] Technical solution: The multi-RIS indoor positioning method based on depth denoising described in this invention includes:

[0007] In indoor scenarios with multipath interference and noise, a fourth-order parallel factor model based on tensors is constructed for the channel state information from non-line-of-sight paths reflected by indoor multi-RIS and spatial scattering points to the receiver.

[0008] To address the issue of a small number of antennas at the transceiver end, an antenna rearrangement technique is used to reconstruct the channel state information tensor model at the receiver end to satisfy the decomposition uniqueness condition.

[0009] To address the problem of severe noise interference in indoor environments, the denoising capabilities of deep learning architecture are utilized to suppress noise contained in the received signal.

[0010] An optimized quadlinear alternating least squares algorithm is used to decompose the tensor, and channel parameters are estimated using subspace-based and correlation-based methods.

[0011] Using the estimated channel parameters and spatial geometric constraints, a search-free spatial node localization method based on geometric relationships is used to estimate the user's position and azimuth angle and map the indoor environment.

[0012] Furthermore, in indoor scenarios with multipath interference and noise, a fourth-order parallel factor model based on tensors is constructed for the channel state information from non-line-of-sight paths reflected by multiple indoor RIS and scattering points in space to the receiver. Specifically, this includes:

[0013] Consider a commercial Wi-Fi-based multiple-input multiple-output (MIMO) orthogonal frequency division multiplexing (OFDM) indoor positioning system, where the number of receiving antennas at the Wi-Fi access point is... And location It is known that the number of transmit antennas of the user equipment (UE) is And azimuth angle With position Unknown. Assume multiple RIS are distributed throughout the indoor environment, each RIS equipped with... The first reflective element. The RIS position corresponding to each path can be represented as , Assume the access point, UE, and multiple RIS (Radio Router Arrays) all utilize a uniform linear array. For simplicity, all antenna elements are arranged at half-wavelength intervals. Assume a severely obstructed indoor scenario where the line-of-sight path is blocked, and the distributed RIS provide a reliable connection between the UE and the Wi-Fi access point. In a real-world deployment, assume each RIS is pre-installed with a specific codebook, whose corresponding codebook matrix is... The passive beamforming vector can be expressed as Among them, the first The path corresponds to the first The first codeword of RIS The phase of each element can be represented , This indicates the number of codebook entries. Assume there are a total of [number] entries between the Wi-Fi access point and the UE. Paths, containing The path via the corresponding RIS single reflection and has The path is generated by reflection from the scattering point. Therefore, if the first... Beamforming code is applied at one RIS location. The overall channel between the WIFI access point and the UE is in the 1st... On a subcarrier, it can be represented as:

[0014]

[0015] in , and . No. The RIS adopts the first The phase control matrix corresponding to each encoding can be expressed as follows: ,in This represents the diagonalization operation of a vector. Furthermore, Indicates the first Subcarrier Zero-mean additive white noise between the access point and the UE under each RIS codeword.

[0016] For a non-line-of-sight path via the scattering point, the access point -SP on the m-th subcarrier ( )-UE channel It can be modeled as:

[0017]

[0018] in , , and These are respectively represented as scattering point index and access point -SP ( - Complex path gain on the UE path, arrival time and subcarrier spacing. This represents the conjugate transpose operation. The steering vectors at the receiver and transmitter are represented as follows:

[0019]

[0020]

[0021] Among them, the access point-SP ( The arrival angle and departure angle on the UE path can be expressed as follows: and , This is represented as a transpose operation. It assumes that automatic frequency control and clock synchronization are applied at the access point and the UE, effectively eliminating the effects of clock skew.

[0022] For a cascaded path via multiple RIS reflections, RIS ( Access Point Channel With UE-RIS ( Channel On the m-th subcarrier, it can be represented as follows:

[0023]

[0024]

[0025] in, and The steering vector of RIS can be represented as:

[0026]

[0027] in, RIS ( The departure angle, arrival angle, arrival time, and complex path gain of the access point path are respectively expressed as: , , and UE-RIS ( The departure angle, arrival angle, arrival time, and complex path gain of the path are expressed as follows: , , and Assuming the channel under consideration and The primary focus is on the line-of-sight path, which is commonly seen in high-frequency scenarios.

[0028] To simplify the subsequent analysis process, The representation can be expanded as follows:

[0029]

[0030] in and These represent the overall gain and overall delay of the cascaded channel, respectively. Therefore, when using the first... When the RIS codeword is used, the first The transmitting antenna to the first The receiving antenna is at the first The channel state information on each subcarrier can be represented as:

[0031]

[0032] in Represents the fourth-order tensor. One element, Indicates effective noise reception. This is represented as following a mean of 0 and a variance of . Gaussian distribution, .also, ,in This is represented as the steering vector of the cascade angle. For cascaded channels, , , and Indicates cascaded angles. For non-line-of-sight channels reflected from scattering points, , , and , .

[0033] Furthermore, the mathematical form of the receiver channel state information tensor can be expressed as:

[0034]

[0035] in , Indicates the outer product. For the noise tensor at the receiving end, . No. The guidance vectors for the arrival time, arrival angle, departure angle, and cascading angle of a path can be represented as follows:

[0036]

[0037]

[0038]

[0039]

[0040] in, , and .

[0041] Furthermore, to address the issue of a small number of antennas at the transceiver end, an antenna rearrangement technique is employed to reconstruct the channel state information tensor model at the receiver to satisfy the decomposition uniqueness condition. Specifically, this includes:

[0042] In multiple-input multiple-output (MIMO) orthogonal frequency division multiplexing (OFDM) systems, subcarrier resources are utilized to enhance spatial resources. Specifically, the signals received by Wi-Fi access points are rearranged to utilize frequency resources to construct virtual antenna array elements. Assuming a total number of... Extracting the number of subcarriers is and The subcarriers are used to expand the number of antennas at the transmitting and receiving ends, respectively. Therefore, the steering vector after using a virtual antenna array can be expressed as:

[0043]

[0044]

[0045]

[0046] in Representing the Kronecker product, the reconstructed fourth-order tensor model of the channel state information can be expressed as:

[0047]

[0048] in , This represents the noise tensor after rearrangement. Furthermore, the tensor... The pattern expansion can be represented as:

[0049]

[0050]

[0051]

[0052]

[0053] The factor matrices corresponding to the fourth-order tensor can be expressed as follows: , , and .

[0054] Furthermore, to address the severe noise interference in indoor environments, the denoising capabilities of deep learning architecture are utilized to suppress noise contained in the received signal, specifically including:

[0055] Consider using a tensor-based complex depth residual convolutional neural network (T-CDRCNN). This T-CDRCNN model consists of one input layer, seven hidden layers, and one output layer. First, the input layer uses 64 convolutional filters, followed by rectified linear units (ReLU) activation to extract 64 feature representations. The kernel convolution size is... Secondly, in the 7 hidden layers, each hidden layer uses 64 elements of size [missing information]. A filter is applied, and batch normalization is inserted between the convolution and activation functions to accelerate convergence and improve denoising efficiency. Finally, a single 3×3×64 convolution is used to reconstruct the output channel matrix. Meanwhile, the channel state information receive tensor and the noise tensor inherently possess complex-valued structures. To utilize... and The correlation between the real and imaginary parts of the data, which T-CDRCNN uses for complex convolution operations, can be represented as:

[0056]

[0057] in, This represents the complex weights of the convolutional network. and They are respectively represented as The real and imaginary parts, and These represent the operations of extracting the real part and the imaginary part, respectively. This represents the complex data values ​​input to the convolutional network. and They are respectively represented as The real and imaginary parts of the integer part. Furthermore, to handle complex data, consider replacing the standard ReLU with a complex ReLU activation function, which can be expressed as:

[0058]

[0059] in, , The T-CDRCNN model is trained for denoising tasks based on residual learning, and the training objective is to minimize the following loss function:

[0060]

[0061] in, This indicates the operation of calculating the Frobenius norm. These are the complex weighting parameters in the T-CDRCNN model. This represents the total number of training data. This is represented as the index value of the training data. Furthermore, and Represent tensors respectively and The pattern - Expand, among which Residual mapping This represents the correspondence between the expansion of a noise tensor and its corresponding expansion. The T-CDRCNN model learns this mapping by minimizing the loss function and continuously optimizing its weight parameters. This enables the denoising of channel state information data.

[0062] After completing the training task of T-CDRCNN, the network realizes the noisy channel state information tensor. With the estimated noise Learning the mapping relationship between them, i.e. Therefore, the denoised channel state information tensor can be represented as:

[0063]

[0064] in, This represents the difference between the output noise of T-CDRCNN and the actual noise.

[0065] Furthermore, an optimized four-linear alternating least squares algorithm is used to decompose the tensor, and channel parameters are estimated using subspace-based and correlation-based methods, specifically including:

[0066] The multi-parameter estimation problem is reformulated as a fourth-order low-rank tensor decomposition problem:

[0067]

[0068] in , , and Representing the factor matrix , , and No. Column elements, Represents the loading matrix The estimated value.

[0069] An improved QALS algorithm is designed to address the fourth-order low-rank tensor fitting problem. First, the initial parameters are estimated using the ESPRIT algorithm (Estimating Signal Parameters with Rotation Invariance). Then, leveraging the rotation invariance of the factor matrix, these parameters are used to reconstruct the initialized factor matrix. .

[0070] Furthermore, the original problem is divided into four subproblems for optimization using a four-linear alternating least squares algorithm, and each factor matrix is ​​updated iteratively until convergence is achieved.

[0071]

[0072]

[0073]

[0074]

[0075] in Represented as the Khatri-Rao product, Indicates the index value of the iteration. , , and Representing the loading matrix respectively , , and In the The result of the next iteration , , and They represent , , and In the The results from the next iteration are then used. Subsequently, using the results obtained from tensor decomposition, a subspace-based method is employed to extract key channel parameters related to terminal localization and mapping. Furthermore, a spatial spectrum is constructed using the projection matrix. Based on the estimated factor matrix, the parameters are then estimated using a spectral peak search method, i.e.:

[0076]

[0077] in Represents the identity matrix. This represents the parameter estimate. , express The corresponding actual value.

[0078] In addition, a correlation-based method is used to analyze the parameters. To make an estimate, that is:

[0079]

[0080] Furthermore, utilizing the estimated channel parameters and spatial geometric constraints, a search-free spatial node localization method based on geometric relationships is employed to estimate the user's position and azimuth angle, as well as map the indoor environment. Specifically, this includes:

[0081] UE-SP ( The geometric relationship between the UE, SP, and WIFI access point in the access point path is represented as follows:

[0082]

[0083]

[0084]

[0085] UE-RIS ( The geometric relationship between the UE, SP, and WIFI access point in the access point path is represented as follows:

[0086]

[0087]

[0088]

[0089]

[0090]

[0091] The above angle parameters are subject to the following constraints: 1) (and 2) Furthermore, under the far-field communication assumption, the number of RIS elements must satisfy the following constraints: 1) (and 2) .in, This represents the spacing between adjacent antennas. This is expressed as wavelength. Therefore, this additional constraint can be expressed as... .

[0092] The estimated channel parameters can be expressed as: The UE angle and position estimation problem, as well as the environment mapping problem, can be expressed as a maximum likelihood estimation problem, i.e.:

[0093]

[0094] Using the given equations and the estimated angle parameters for each path, an overdetermined equation for the unknown UE azimuth angle can be derived and solved using the least squares method. Furthermore, our goal is to utilize the estimated azimuth angle... , Location known and the estimated path parameters To obtain the location of the UE and SPs. The UE location can be represented by various paths:

[0095]

[0096] The angle vectors of the receiver and transmitter can be expressed as follows: and . Represented as the corresponding number The location weight coefficients of each path. To simplify the subsequent representation, the UE location expression can be rearranged as follows:

[0097]

[0098] in, and Therefore, the UE localization problem can be further modeled as the intersection problem of a system of linear equations. Accordingly, the cost function is constructed as the sum of distances along all propagation paths:

[0099]

[0100] in express The weights corresponding to each path depend on the signal-to-noise ratio of that path. This is achieved by minimizing the cost function. Thus, its least squares solution is obtained:

[0101]

[0102] in This represents the inverse operation. After obtaining the estimated user location... Then, the location of the spatial scattering point It can be by and The intersection points represent, where Therefore, the first Estimated location of scattering point corresponding to each path It can be expressed by the following formula:

[0103]

[0104] in and .

[0105] Beneficial Effects: Compared with existing technologies, its main advantages are: This invention can accurately estimate the UE position angle and map the indoor environment in indoor environments with limited antennas, complex multipaths, and noise interference; it effectively removes noise interference from the received signal using a CNN-based T-CDRCNN model, thus improving the accuracy and robustness of the indoor positioning system; it utilizes a virtual antenna rearrangement method based on a fourth-order channel state information tensor, achieving accurate estimation of channel parameters and target position azimuth angle without the need for additional equipment; and it uses multiple RIS for indoor positioning assistance, improving spatial resolution to achieve more accurate indoor positioning results. The advantages and methods of this invention can be further understood through the following detailed description and accompanying drawings. Attached Figure Description

[0106] Figure 1 This is a flowchart of a multi-RIS indoor positioning method based on depth denoising according to the present invention;

[0107] Figure 2 This is a schematic diagram of the multi-RIS indoor positioning system of the present invention;

[0108] Figure 3 To provide the present invention with different RIS codebook numbers and the number of subcarriers Below is a performance graph showing the relationship between the root mean square error of arrival (RMSE) and signal-to-noise ratio (SNR) for angle of arrival estimation, compared to existing methods.

[0109] Figure 4 To provide the present invention with different RIS codebook numbers and the number of subcarriers Below is a performance graph showing the relationship between RMSE and SNR for arrival time estimation and existing methods;

[0110] Figure 5 To provide the present invention with different RIS codebook numbers and the number of subcarriers Below is a performance graph comparing the RMSE and SNR of the departure angle estimation method with existing methods;

[0111] Figure 6 To provide the present invention with different RIS codebook numbers and the number of subcarriers Below is a performance graph showing the relationship between RMSE and SNR of cascaded angle estimation methods compared to existing methods;

[0112] Figure 7 To provide the present invention with different RIS codebook numbers and the number of subcarriers Below is a performance graph showing the relationship between RMSE and SNR for user location estimation and existing methods;

[0113] Figure 8 To provide the present invention with different RIS codebook numbers and the number of subcarriers Below is a performance graph comparing the RMSE and SNR of scattering point location estimation with existing methods;

[0114] Figure 9 To provide the present invention with different RIS codebook numbers and the number of subcarriers Below is a performance graph showing the relationship between RMSE and SNR for user azimuth estimation using existing methods. Detailed Implementation

[0115] The preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby providing a clearer and more explicit definition of the scope of protection of the present invention.

[0116] Figure 2 This is a schematic diagram of the multi-RIS indoor positioning system of the present invention, as shown below. Figure 2 The uplink system for indoor user positioning using multiple RIS-assisted WIFI access points is shown. The signal sent by the user equipment reaches the WIFI terminal through reflection from multiple passive RIS devices distributed in space and scattering points in space. The WIFI location is known, while the scattering points, user location, and user azimuth are unknown.

[0117] Implementation Example 1

[0118] Please see Figure 3 , Figure 4 , Figure 5 and Figure 6 These four figures illustrate the invention in different RIS codebooks. and the number of subcarriers Below is a performance graph showing the relationship between RMSE and SNR for each channel parameter estimation method compared to existing target sensing methods. The parameters are set as follows: , , The size of the node distribution space is Simulation results show that as the number of RIS codebooks increases... With the increase in the number of subcarriers and the improvement in spatial resolution, the RMSE of the channel parameters estimated by the proposed algorithm and existing algorithms—angle of arrival, departure angle, time of arrival, and concatenation angle—is all reduced. Furthermore, with the increase in the number of subcarriers... With the increase of SNR, the RMSE of the channel parameter estimation structures of both the proposed and existing methods is reduced. Furthermore, as SNR increases, the RMSE of both the proposed and existing algorithms decreases, and the RMSE curves of the proposed method are all below those of the existing methods, further demonstrating that the parameter extraction performance of the proposed algorithm has a significant advantage over existing algorithms.

[0119] Implementation Example 2

[0120] Please see Figure 7 , Figure 8 and Figure 9 These three figures illustrate the invention in different RIS codebooks. and the number of subcarriers Below is a performance graph comparing the RMSE and SNR of existing target perception methods for estimating the azimuth angle of the user target location and the location of scattering points in space. Parameters are set as follows: , , The size of the node distribution space is Simulation results show that as the number of subcarriers increases... The proposed method, compared to existing methods, shows a decrease in the RMSE for user location azimuth estimation, indicating that increased spectral efficiency contributes to accurate target localization and azimuth awareness. Furthermore, with the increase in RIS codebook size... With the increase of SNR, the accuracy of target localization and environment perception of the algorithm is improved due to the enhanced spatial resolution capability of the localization system. As SNR increases, the RMSE of the proposed method decreases accordingly, while always remaining below the RMSE curve of existing methods, indicating that the proposed method has better robustness in localization and environment mapping.

[0121] In summary, this invention proposes a deep denoising-based multi-RIS indoor positioning method. It constructs a fourth-order tensor model from the channel state information reflected from multiple indoor RIS points and scattering points to the receiver. This tensor model is then rearranged using antenna rearrangement techniques. Furthermore, a deep learning architecture is employed to remove noise from the received signal. Subsequently, an optimized quadlinear alternating least squares algorithm is used to extract channel parameters. Finally, a search-free spatial positioning method is used to estimate the user's location and azimuth angle, as well as map the indoor environment. Therefore, the deep denoising-based multi-RIS indoor positioning method proposed in this invention exhibits higher accuracy and robustness compared to comparative algorithms, better meeting the needs of practical communication scenarios.

[0122] The above description of the embodiments is only intended to help understand the method and main idea of ​​the present invention. The content of this specification should not be construed as limiting the scope of the invention; therefore, the scope of protection of the present invention should be determined by the appended claims.

Claims

1. A multi-RIS indoor positioning method based on depth denoising, characterized in that... The method includes: In indoor scenarios with multipath interference and noise, a fourth-order parallel factor model based on tensors is constructed for the channel state information from non-line-of-sight paths reflected by multiple indoor RIS points and scattering points in space to the receiver. Specifically, this includes considering a commercial WIFI-based MIMO-Orthogonal Frequency Division Multiplexing indoor positioning system, where the number of receiving antennas at the WIFI access point is... And location It is known that the number of transmit antennas of the user equipment is And azimuth angle With position It is unknown, assuming that multiple RIS are distributed in the indoor environment, and each RIS is equipped with The nth reflective element, of which the nth reflective element is the nth reflective element. The RIS position corresponding to each path can be represented as , Assuming the access point, UE, and multiple RIS all employ a uniform linear array, and for simplicity, all antenna elements are arranged at half-wavelength intervals, and that in a severely obstructed indoor scenario where the line-of-sight path is blocked, the distributed RIS provides a reliable connection between the UE and the Wi-Fi access point. In actual deployment, it is assumed that each RIS is pre-installed with a specific codebook, and its corresponding codebook matrix is ​​as follows: The passive beamforming vector can be expressed as , among which, the The path corresponds to the first The first codeword of RIS The phase of each element can be represented , This represents the number of codebook entries. Assume there are a total of [number] entries between the Wi-Fi access point and the UE. Paths, containing The path via the corresponding RIS single reflection and has The path is generated by reflection from the scattering point; therefore, if the first path is... Beamforming code is applied at one RIS location. The overall channel between the WIFI access point and the UE is in the 1st... On the subcarrier, it can be represented as ， in , and , No. The RIS adopts the first The phase control matrix corresponding to each encoding can be expressed as follows: ,in This represents the diagonalization operation of a vector, in addition... Indicates the first Subcarrier Zero-mean additive white noise between the access point and the UE under the RIS codeword, for a non-line-of-sight path via the scattering point, the th Access Point on Subcarrier - SP -UE channel Modeling as ， in , , and These represent the scattering point index and the access point -SP, respectively. - Complex path gain on the UE path, arrival time and subcarrier spacing The conjugate transpose operation is represented by the steering vectors at the receiver and transmitter, respectively. ， ， Among them, access point-SP The arrival angle and departure angle on the UE path can be expressed as follows: and , Represented as a transpose operation, for a cascaded path via multiple RIS reflections, RIS -Access Point Channel with UE-RIS Channel On the m-th subcarrier, it can be represented as follows: ， ， in and The steering vector of RIS can be represented as ， in RIS - The departure angle, arrival angle, arrival time, and complex path gain of the access point path are respectively expressed as: , , and UE-RIS The departure angle, arrival angle, arrival time, and complex path gain of the path are respectively expressed as: , , and , The representation can be expanded as ， in and These are respectively represented as the overall gain and overall delay of the cascaded channel, when the first... When the RIS codeword is used, the first The transmitting antenna to the first The receiving antenna is at the first The channel state information on each subcarrier can be represented as follows: ， in Representing a fourth-order tensor The Middle One element, Indicates effective noise reception. This is represented as following a mean of 0 and a variance of . Gaussian distribution, ,also, , The steering vector, represented as the cascade angle, is used for cascaded channels. , , and Indicates cascaded angles. For non-line-of-sight channels reflected from scattering points, , , and , The mathematical form of the receiver channel state information tensor can be expressed as: ， in , Indicates the outer product. For the noise tensor at the receiving end, , No. The orientation vectors for the arrival time, arrival angle, departure angle, and cascading angle of a path can be expressed as follows: ， ， ， ， in , and , To address the issue of a limited number of antennas at the transceiver end, an antenna rearrangement technique is employed to reconstruct the channel state information tensor model at the receiver, ensuring the uniqueness of the decomposition. Specifically, this includes assuming a total number of antennas... Extracting the number of subcarriers is and The subcarriers are used to expand the number of antennas at the transmitting and receiving ends, respectively. Therefore, the steering vector after using a virtual antenna array can be expressed as... ， ， ， in Representing the Kronecker product, the reconstructed fourth-order tensor model of the channel state information can be expressed as: ， in , This represents the noise tensor after rearrangement. The pattern expansion can be represented as ， ， ， ， in , , and , To address the severe noise interference in indoor environments, this paper utilizes the denoising capabilities of deep learning architectures to suppress noise in the received signal. Specifically, it employs a T-CDRCNN model, consisting of one input layer, seven hidden layers, and one output layer. First, the input layer uses 64 convolutional filters, followed by ReLU activation to extract 64 feature representations. The kernel convolution size is... Secondly, in the 7 hidden layers, each hidden layer uses 64 elements of size 1. The filter is optimized, and batch normalization is inserted between the convolution function and the activation function to accelerate convergence and improve denoising efficiency. Finally, a single 3×3×64 convolution is used to reconstruct the output channel matrix. Meanwhile, in order to utilize... and The correlation between the real and imaginary parts of the data, T-CDRCNN uses complex-valued convolution operations, which can be represented as: ， in This represents the complex weights of the convolutional network. and They are respectively represented as The real and imaginary parts, and These represent the operations of extracting the real part and the imaginary part, respectively. This represents the complex data values ​​input to the convolutional network. and They are respectively represented as Given the real and imaginary parts of , consider replacing the standard ReLU with a complex ReLU activation function, which can be expressed as: ， in , The T-CDRCNN model is trained for the denoising task based on residual learning, and the training objective is to minimize the following loss function. ， in, This indicates the operation of calculating the Frobenius norm. These are the complex weighting parameters in the T-CDRCNN model. This represents the total number of training data. Represented as the index value of the training data. and Represent tensors respectively and The pattern - Expand, among which Residual mapping To represent the correspondence between the expansion of a noise tensor and its corresponding expansion, the T-CDRCNN model continuously optimizes its weight parameters by minimizing the loss function. Let's learn about mapping relationships This achieves denoising of the channel state information data, and the denoised channel state information tensor can be represented as: ， in, This represents the difference between the output noise of T-CDRCNN and the actual noise. An optimized quadlinear alternating least squares algorithm is employed to decompose the tensor, and channel parameters are estimated using subspace-based and correlation-based methods. Specifically, the multi-parameter estimation problem is reformulated as a fourth-order low-rank tensor decomposition problem. ， in , , and Representing the factor matrix , , and No. Column elements, Represents the loading matrix The estimated values ​​are obtained by using the ESPRIT algorithm for initial parameter estimation, and the initial factor matrix is ​​reconstructed using the rotation invariance of the factor matrix. The original problem is divided into four subproblems using a four-linear alternating least squares algorithm for optimization. Each factor matrix is ​​updated iteratively until convergence. ， ， ， ， in Represented as the Khatri-Rao product, Indicates the index value of the iteration. , , and Representing the loading matrix respectively , , and In the The result of the next iteration , , and They represent , , and In the The results from the next iteration, obtained using tensor decomposition, are used to extract key channel parameters related to terminal positioning and mapping using a subspace-based method, and a spatial spectrum is constructed using the projection matrix. The parameters are estimated using a spectral peak search method. ， in Represents the identity matrix. This represents the parameter estimate. , express The corresponding true values; in addition, a correlation-based method is used to analyze the parameters. Make an estimate ， Using the estimated channel parameters and spatial geometric constraints, a search-free spatial node localization method based on geometric relationships is employed to estimate the user's position and azimuth angle, as well as map the indoor environment. Specifically, this includes: UE-SP - The geometric relationship between the UE, SP, and WIFI access point path is represented as follows: ， ， ， UE-RIS - The geometric relationship between the UE, SP, and WIFI access point path is represented as follows: ， ， ， ， ， The angle parameter is constrained. and Under the far-field assumption, the additional constraint satisfied by the number of RIS elements is expressed as follows: ,in This represents the spacing between adjacent antennas. Expressed as wavelength, the estimated channel parameters can be expressed as... The UE angle position estimation problem and the environment mapping problem can be represented as a maximum likelihood estimation problem. ， Using the given equations and combining them with the estimated angle parameters for each path, an overdetermined equation for the unknown UE azimuth angle can be derived, thus allowing the UE's position to be represented by each path. ， The angle vectors of the receiver and transmitter can be expressed as follows: and , Represented as the corresponding number The location weight coefficients of each path, after rearranging the UE location expression, can be further expressed as: ， in and Therefore, the UE localization problem can be further modeled as the intersection problem of a system of linear equations, and the cost function is constructed as the sum of distances along all propagation paths. ， in express The weights corresponding to each path are used to obtain the least squares solution of the cost function. ， in This represents the inverse operation, used to obtain the estimated user location. Then, the location of the spatial scattering point It can be by and The intersection points represent, where , No. Estimated location of scattering point corresponding to each path It can be expressed by the following formula. ， in and .

Images