Reconfigurable intelligent surface-assisted terahertz indoor positioning algorithm based on dictionary optimization

Abstract

The invention provides a dictionary optimization-based reconfigurable intelligent surface-assisted terahertz indoor positioning algorithm. The dictionary optimization-based reconfigurable intelligent surface-assisted terahertz indoor positioning algorithm comprises the following steps of: 1, receiving a signal randomly reflected by a reconfigurable intelligent surface (RIS) at an initial stage by user equipment (UE); step 2, obtaining coarse estimation channel state information (CSI) based on channel estimation of dictionary optimization and compressed sensing (CS); 3, designing a reconfigurable intelligent surface RIS reflection phase, and outputting an optimal beam direction; step 4, the UE performs channel estimation again, and outputs optimized channel state information (CSI); and step 5, carrying out position calculation and optimization based on maximum likelihood estimation (MLE). The scheme provides an effective method for realizing high-precision channel modeling and indoor positioning in a terahertz near-field environment, and has a wide application prospect.

Description

Technical Field

[0001] The present application relates to the field of indoor positioning technology, and in particular to a reconfigurable intelligent surface-assisted terahertz indoor positioning algorithm based on dictionary optimization. Background Art

[0002] With the rapid development of wireless communication technology, millimeter wave frequency bands have been widely used in 5G communications, significantly improving data transmission rates. However, with the rapid growth of communication needs, the limitations of 5G millimeter wave frequency bands have gradually emerged, especially in complex environments. In order to meet the needs of future 6G and subsequent technologies for high data rates and high positioning accuracy, researchers have gradually turned their attention to terahertz band communication technology. Due to its ultra-wideband characteristics, terahertz communication can provide bandwidth resources far exceeding millimeter wave communication, and has better spatial resolution at higher frequencies, making it an important technology for future indoor positioning services and wireless communication systems.

[0003] Terahertz communication not only has advantages in data transmission, but also plays a key role in high-precision positioning services. Due to the short wavelength characteristics of the terahertz band, it has the ability to estimate angles with high precision, especially in future scenarios such as indoor navigation, virtual reality, and augmented reality, providing more accurate location information. In addition, the introduction of reconfigurable intelligent surfaces (RIS) provides an effective solution to signal coverage, reflection, and multipath interference problems. RIS is an emerging technology that can optimize the signal propagation path by dynamically adjusting the electromagnetic properties of the surface, thereby improving positioning and communication performance in complex indoor environments.

[0004] Although RIS has made some progress in the research of millimeter wave frequency band, it still faces many challenges when applied in the higher frequency terahertz band. Due to its propagation characteristics, the signal in the terahertz band is easily affected by the beam splitting effect (BSE) and the near-field spherical wave effect. The beam splitting effect refers to the fact that different frequency components of the terahertz broadband signal will propagate along different spatial paths, resulting in signal mismatch and positioning errors; while the spherical wave effect makes the propagation of the signal no longer conform to the far-field assumption under near-field conditions, especially the complex phase changes, which increases the difficulty of channel estimation.

[0005] Existing RIS-based millimeter wave positioning technology usually assumes that signal propagation meets far-field conditions when dealing with these problems. However, in the terahertz frequency band, this assumption is often no longer valid. Especially in indoor scenes, the frequent occurrence of signal reflection, diffraction and multipath effects makes traditional beam training and channel estimation methods unable to cope with the complex propagation characteristics in the terahertz communication environment. Therefore, how to design a RIS-assisted positioning system suitable for the terahertz frequency band and solve the challenges of beam splitting effect and spherical wave propagation has become a difficult problem that current technology urgently needs to break through.

[0006] Defects and shortcomings of the existing technology:

[0007] Most of the existing RIS-assisted positioning systems are applied to millimeter wave communications, which rely on far-field beamforming and simple geometric propagation models. In terahertz scenarios, these systems have the following shortcomings:

[0008] 1. Beam splitting effect: Due to the broadband characteristics of terahertz communication, different frequency components of the signal will propagate along different paths in space, resulting in splitting and mismatching of the received signal. This effect significantly reduces the accuracy of existing beamforming and beam training algorithms under high-frequency broadband conditions. Especially for positioning services, the beam splitting effect will directly lead to increased errors in angle estimation and distance measurement.

[0009] 2. Inadequate processing of near-field effects: In the terahertz frequency band, due to the short propagation distance of the signal, the traditional far-field assumption is no longer applicable. The propagation characteristics of the signal under near-field conditions are more complex, and the signal phase front appears as a spherical wave rather than a plane wave in the far field. This characteristic makes it difficult for the existing far-field beam training and channel estimation algorithms to accurately capture the phase changes of the signal in terahertz near-field communications, resulting in a decrease in the positioning accuracy of the system.

[0010] 3. Sparse channel characteristics: The terahertz channel has a significant sparse characteristic, and the traditional least squares (LS) channel estimation method cannot effectively utilize this characteristic. The LS method relies on a large amount of training resources and is highly sensitive to complex channel models. It is easy to produce large estimation errors in the terahertz frequency band, reducing the robustness and accuracy of the system.

[0011] 4. High computational complexity: In the terahertz scenario, the configuration of RIS and the beamforming algorithm need to process a large number of frequency components and spatial dimensions, so the computational complexity increases significantly. The existing system cannot effectively cope with the computational requirements in the terahertz frequency band, resulting in low processing efficiency in practical applications. Summary of the invention

[0012] In view of this, the present application provides a reconfigurable smart surface-assisted terahertz indoor positioning algorithm based on dictionary optimization to improve the robustness and positioning accuracy of the system.

[0013] To achieve the above objectives, the technical solutions adopted in this application are as follows:

[0014] Reconfigurable smart surface-assisted terahertz indoor positioning algorithm based on dictionary optimization, including:

[0015] Step 1: The user equipment UE receives the signal randomly reflected by the reconfigurable smart surface RIS in the initial stage;

[0016] Step 2: Channel estimation based on dictionary optimization and compressed sensing CS to obtain coarse estimated channel state information CSI;

[0017] Step 3: Design the reflection phase of the reconfigurable smart surface RIS and output the optimal beam direction;

[0018] Step 4: The user equipment UE performs channel estimation again and outputs optimized channel state information CSI;

[0019] Step 5: Position solution and optimization based on maximum likelihood estimation (MLE).

[0020] Furthermore, before step 1, the method further includes establishing a channel model of the reconfigurable intelligent surface-assisted terahertz indoor positioning system, specifically including determining a channel matrix H from the base station to the RIS. B,R , the channel matrix H from RIS to user equipment R,U and the phase control matrix Θ[n] of RIS on the nth subcarrier;

[0021] Furthermore, the channel matrix H from the base station to the RIS B,R It is expressed as:

[0022]

[0023] Among them, N B is the number of base station antennas, N B is the number of reflection units in RIS, H B,R [n] is the channel matrix between the base station and the RIS on the nth subcarrier; N B and N R are the number of antennas of the base station and RIS respectively; L B,R is the number of multipaths between the base station and the RIS; α l,B,R [n] is the complex gain of the lth path at the nth subcarrier, which depends on the path loss and phase variation; and denote the response vectors of RIS and base station antenna array respectively, which depend on the incident angle exit angle and path distance

[0024] The channel matrix H from RIS to user equipment R,U , expressed as:

[0025]

[0026] Among them, H R,U [n] is the channel matrix between RIS and user equipment on the nth subcarrier; N U is the number of antennas on the user equipment; L R,U is the number of multipaths between RIS and user equipment; α l,R,U [n] is the complex gain of the lth path on the nth subcarrier, which depends on the path distance and environmental influences; and denote the response vectors of the user equipment and the RIS antenna array, respectively, which depend on the signal departure angle Angle of arrival and path distance

[0027] Further, the phase control matrix Θ[n] of the RIS on the nth subcarrier is expressed as:

[0028]

[0029] Where Θ[n] is the phase control matrix of RIS on the nth subcarrier, β i [n] is the reflection coefficient of the i-th RIS unit on the n-th subcarrier, θ i [n] is the phase offset of the i-th RIS unit.

[0030] Furthermore, the specific method of obtaining coarse estimated channel state information CSI based on dictionary optimized channel estimation in step 2 includes:

[0031] Step 2.1: Receive RIS reflected signal y, where y = H RIS ·H BS ·s+n, the H RIS represents the channel matrix of RIS, H BS represents the channel matrix of the base station, s represents the signal sent by the base station, and n represents the noise;

[0032] Step 2.2: The first antenna subarray is estimated by orthogonal matching pursuit (OMP) to obtain the first sparse coefficient vector a1, and Θ ovs represents the initial complete redundant dictionary, Ψ is the transformation basis of compressed sensing technology;

[0033] Step 2.3: Generate support set supp(a1);

[0034] Step 2.4: Based on the support set, the initial complete redundant dictionary Θ ovs Perform optimization and reduction, retaining only the entries corresponding to the support set, to form an optimized dictionary Θ DO , where Θ DO =Θ OVS [:,support(a1)];

[0035] Step 2.5: Use the simultaneous orthogonal matching pursuit SOMP algorithm to estimate the channel and obtain the optimal solution of the channel matrix Extract channel state information CSI, and

[0036] Step 2.6: Output channel state information CSI and

[0037] Furthermore, the simultaneous orthogonal matching pursuit SOMP algorithm specifically includes:

[0038] (1) First calculate the dictionary column index i* that is most relevant to the current residual:

[0039]

[0040] The T represents the transpose of the matrix;

[0041] (2) Then, update the sparse representation coefficient α DO And the residual r:

[0042]

[0043] r=y-ΨΘ DO α DO (4.21)

[0044] (3) The above process (1)-(2) is iterated continuously until the residual is small enough or the preset number of iterations is reached, and the sparse representation coefficient α is finally extracted. DO Corresponding to a specific θ AOA ,θ AOD and τ TOA Specifically, for θ AOA and θ AOD , assuming that the dictionary entry Θ DO The angle corresponding to [:,i] is θ i , then the extracted θ AOA and θ AOD They are:

[0045] θ AOA =θ Rx ,θ AOD =θ Tx(4.22)

[0046] Among them, θ Rx and θ Tx are the receiving and sending angles from RIS to UE respectively; for τ TOA , dictionary entry Θ DO [:,i] also contains the time delay information τ i , the extracted τ TOA for:

[0047] τ TOA =τ i

[0048] The θ AOD represents the departure angle from RIS to UE, θ AOA represents the arrival angle from RIS to UE, τ TOA Indicates the arrival time from RIS to UE.

[0049] Furthermore, the specific method of designing the reconfigurable intelligent surface RIS reflection phase and outputting the optimal beam direction in step 3 includes:

[0050] Step 3.1: Obtain channel state information CSI (θ AOD ,θ AOA ,τ TOA );

[0051] Step 3.2: Generate initial codebook initial =f(θ1,θ2,…,θ k );

[0052] Step 3.3: Select the optimal initial beam direction θ opt , W represents the beamforming vector or weight vector, represents the channel estimation matrix, H r represents the channel matrix of the reflection path, k represents the subcarrier index, and t represents the time index, which is used to describe the channel state that changes over time. represents the channel matrix generated by the Kth subcarrier at time node t;

[0053] Step 3.4: Generate high-level codebook high =phasecorrection(Codebook initial ), The φ corrected ( d ) represents the angle obtained after phase correction based on the initial codebook.

[0054] Step 3.5: Beam training based on power distribution pattern PDP;

[0055] Step 3.6: Confirm whether the current optimal beam direction is consistent with θ PDP Match, if not, return to step 3.3, otherwise go to step 3.7; the θ PDP is the UE beam direction calculated based on the power distribution pattern;

[0056] Step 3.7: Adjust the reconfigurable smart surface RIS reflection phase Φ so that the RIS reflection beam is accurately directed to the user equipment UE, and where ρ opt The phase parameter representing the optimized beam direction;

[0057] Step 3.8: Output the optimized channel state information CS I.

[0058] Furthermore, in step 3.2, the formula for generating the primary codebook is as follows:

[0059]

[0060] Among them, d n,m is the antenna spacing, λ is the signal wavelength, θ i It is the perspective between the target user and the RIS.

[0061] Furthermore, in step 3.4, the phase change Δφ of the spherical wave s-wave Specifically:

[0062]

[0063] The corrected beam phase can be expressed as:

[0064]

[0065] Where d is the distance the signal travels, R is the radius of curvature of the phase front during beamforming, and λ is the wavelength of the signal.

[0066] The generation formula of the advanced codebook is as follows:

[0067] C s =(A H A) -1 A H G s (4.27)

[0068] Among them, C s is the advanced codebook, and A is the modified primary codebook matrix.

[0069] Furthermore, the specific method of the beam training based on the power distribution mode PDP in step 3.5 is:

[0070] The phase shift matrix of RIS can be expressed as:

[0071]

[0072] Among them, φ i represents the phase shift of RIS, N is the number of reflection units of RIS; for each subcarrier, the power of the user received signal can be expressed as:

[0073]

[0074] Among them, h UE is the UE’s receive antenna array response vector, h RIS are the array response vectors of the RIS; these vectors are based on a typical uniform linear array structure, where the elements in the antenna array are equally spaced and arranged in a straight line. The array response vector h RIS and h UE is obtained by accumulating phase differences, which are related to the position and incident angle (θ AOA / θ AOD ) related.

[0075] In the PDP strategy, the direction of the UE is analyzed and estimated by calculating the distribution pattern of the received power in space. Usually, two adjacent wide beams are set to capture the difference in the power distribution received in space, so that the difference in received power in different directions is calculated by the following formula:

[0076]

[0077] Where j1(φ) and j2(φ) represent the power responses of two adjacent beams received in the direction φ. Due to the beam splitting effect of subcarriers of different frequencies in a broadband system, the direction estimation in space may be offset. However, by analyzing the power ratio ε(φ), the unique θ can be obtained. AOA and θ AOD more precise estimate of .

[0078] The beneficial effects of this application are:

[0079] 1. High-precision channel description: The spherical wave model is introduced to accurately characterize the channel characteristics of the terahertz near-field MIMO system. Compared with the traditional plane wave model, it can better reflect the distance dependence of signal propagation under near-field conditions, thereby improving the accuracy of the channel model.

[0080] 2. Optimize signal propagation and positioning accuracy: Through the design of the RIS phase control matrix, the reflection phase is adjusted to achieve intelligent control of the signal propagation path, significantly improving the system's coverage and positioning accuracy.

[0081] 3. Reduce computational overhead: The channel estimation method based on compressed sensing and dictionary optimization can accurately restore the channel matrix through a small number of pilot symbols, greatly reducing the training overhead required for channel estimation. The optimized dictionary design method further reduces computational complexity and storage requirements.

[0082] 4. Wide application scenarios: The proposed algorithm combines spherical wave and plane wave models, which can effectively process channel characteristics in far-field and near-field environments, and is suitable for wireless communication and positioning systems in 6G and terahertz frequency bands.

[0083] 5. This scheme provides an effective method to achieve high-precision channel modeling and indoor positioning in the terahertz near-field environment, and has broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0085] Figure 1 It is a complete positioning flow chart of the present invention;

[0086] Figure 2 A flow chart of the implementation scheme for channel estimation and channel state information (CSI) extraction based on compressed sensing (CS) and dictionary optimization;

[0087] Figure 3 It is a flow chart of RIS reflection phase design based on power distribution pattern (PDP) and hierarchical codebook design;

[0088] Figure 4 This is a schematic diagram of the PDP-based beam training of the present application. DETAILED DESCRIPTION

[0089] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments.

[0090] like Figure 1As shown, the present application provides a reconfigurable intelligent surface-assisted terahertz indoor positioning algorithm based on dictionary optimization, including:

[0091] Step 1: The user equipment UE receives the signal randomly reflected by the reconfigurable smart surface RIS in the initial stage;

[0092] In the RIS-assisted terahertz near-field system, the propagation characteristics of the signal are not only affected by the angle, but also closely related to the propagation distance. Since the wavelength of the terahertz frequency band signal is extremely short, when the signal propagates between the transmitter and the receiver, the changes in its phase and amplitude will significantly depend on the propagation distance. This dependence is particularly important in the near-field environment, so it is necessary to introduce a spherical wave model to accurately describe the channel characteristics. The traditional plane wave model assumes that the wavefront of the signal is parallel, but in the near-field environment, this assumption is no longer applicable. In order to better characterize the channel characteristics of the terahertz near-field MIMO system, it is necessary to derive the channel models from the base station to the RIS and from the RIS to the UE, and describe how to model each subcarrier in the OFDM system.

[0093] First, consider a typical two-hop NLoS path channel model, which includes the first-hop channel from the base station to the RIS and the second-hop channel from the RIS to the UE. In this system, the channel consists of two parts: the channel matrix H from the base station to the RIS B,R and the channel matrix H from RIS to the user equipment R,U The specific representation of these two channel matrices determines the overall channel characteristics of the system. In the terahertz near-field environment, the propagation distance of the signal will have a significant impact on the channel characteristics, so the channel matrix is ​​represented as a function of distance.

[0094] For the channel from the base station to the RIS, it is assumed that both the base station and the RIS use a uniform linear array (ULA), and the base station has N B antennas, RIS has N R The channel matrix H from the base station to the RIS B,R It can be expressed as:

[0095]

[0096] Among them, H B,R [n] is the channel matrix between the base station and the RIS on the nth subcarrier; N B and N R are the number of antennas of the base station and RIS respectively; L B,R is the number of multipaths between the base station and the RIS; α l,B,R [n] is the complex gain of the lth path at the nth subcarrier, which depends on the path loss and phase variation; and denote the response vectors of RIS and base station antenna array respectively, which depend on the incident angle exit angle and path distance

[0097] Under the spherical wave model, the response vector of the antenna array depends not only on the incident angle, but also on the propagation distance. Specifically, the response vector of the antenna array can be expressed as:

[0098]

[0099] Among them, θ is the incident angle of the signal, r is the propagation distance of the signal, λ is the carrier wavelength, d is the antenna spacing, and N is the number of antennas in the antenna array. This formula describes the phase change of the signal, which not only depends on the incident angle θ of the signal, but is also significantly affected by the signal propagation distance r.

[0100] For the channel matrix H from RIS to user equipment R,U , can be similarly expressed as:

[0101]

[0102] Among them, H R,U [n] is the channel matrix between RIS and user equipment on the nth subcarrier; N U is the number of antennas on the user equipment; L R,U is the number of multipaths between RIS and user equipment; α l,R,U [n] is the complex gain of the lth path on the nth subcarrier, which depends on the path distance and environmental influences; and denote the response vectors of the user equipment and the RIS antenna array, respectively, which depend on the signal departure angle Angle of arrival and path distance

[0103] In order to more accurately describe the signal propagation characteristics in the terahertz near-field channel, the channel gain α l,B,R [n] and α l,R,U [n] It is necessary to consider the effect of signal propagation distance on signal amplitude and phase. According to the free space path loss model, the channel gain can be expressed as:

[0104]

[0105] in, and are the path distances from the base station to the RIS and from the RIS to the UE. These distance factors directly affect the attenuation and phase change of the channel. Here λ is the wavelength of the signal, and the phase term in the formula Describes the phase shift of the signal due to the distance r during propagation.

[0106] In addition, the phase control matrix θ of RIS can be expressed as:

[0107]

[0108] Where Θ[n] is the phase control matrix of RIS on the nth subcarrier, β i [n] is the reflection coefficient of the i-th RIS unit on the n-th subcarrier, θ i [n] is the phase offset of the i-th RIS unit. This matrix controls the reflection path of the signal by adjusting the reflection phase of each RIS unit to optimize the coverage and positioning accuracy of the system. Combining the above parts, the signal matrix Y[n] received by the user equipment on the n-th subcarrier can be expressed by the following formula:

[0109] Y[n]=H R,U [n]Θ[n]H B,R [n]s[n]+n[n](4.6)

[0110] Wherein, Y[n] is the signal matrix received by the user equipment on the nth subcarrier; s[n] is the signal sent by the base station on the nth subcarrier; and n[n] is the noise vector.

[0111] In order to more specifically describe the multipath propagation characteristics of the system and the received signal model, it is assumed that the RIS reflection unit is located at a fixed position in space. In this model, r represents the distance between the reflection unit and the source point p. RIS The Euclidean distance between i is the position of the ith unit of RIS, and ∥pq i ∥ represents the distance from the user equipment to the i-th reflection unit of the RIS. In order to consider the influence of time delay, a model of the downlink UE receiving signal is introduced, in which the time delay of the reflection path is not only related to the path length, but also to the phase control of the RIS unit.

[0112] Based on the above analysis, it can be concluded that the signal y received by the user equipment during the Gth transmission is g for:

[0113]

[0114] Among them, h B,R is a fixed known channel from BS to RIS, Ωg is the phase control matrix of the RIS in the g-th transmission, a(p) is the steering vector from UE to RIS, n g is the noise vector.

[0115] After further derivation, the received signals on multiple subcarriers can be combined and expressed as:

[0116]

[0117] Where H is the combination matrix of all transmission channels, and n is the noise vector after combination. The above formula and derivation describe in detail the channel model of the RIS-assisted positioning system in the terahertz near-field environment. This model not only takes into account the angle dependence and distance dependence of signal propagation, but also fully considers the changes in time delay and channel gain, providing a theoretical basis for the design and optimization of the system.

[0118] Step 2: Channel estimation based on dictionary optimization and compressed sensing CS to obtain coarse estimated channel state information CSI;

[0119] In indoor positioning systems, accurate extraction of CSI is crucial for beamforming and positioning of the system. In the terahertz near-field environment, traditional channel state information extraction methods face challenges due to the sparsity of the channel and the beam splitting effect. Commonly used channel state information extraction methods include Multiple Signal Classification (MUSIC) and Estimation of Signal Parameters via Rotational Invariance Techniques (ESPRIT) methods.

[0120] The MUSIC and ESPRIT algorithms are classic channel state information (CSI) extraction methods, which are often used for angle and time delay estimation. The MUSIC algorithm uses the orthogonality of the noise subspace and the signal subspace to estimate the angle and time delay information by constructing a spectral function. The ESPRIT algorithm is based on the rotational invariance of the array and uses the characteristic equation to simultaneously estimate the angle and time delay. These two algorithms perform well in millimeter wave far-field positioning, but in terahertz near-field conditions, CSI extraction faces challenges due to the short wavelength and obvious beam splitting effect. Under these conditions, the MUSI and ESPRIT algorithms skip the channel estimation step of OMP and directly match the received signal with the complete dictionary matrix to extract the angle and time delay information in the CSI. Without passing through OMP, the received signal is directly stacked into a channel matrix, which is actually a stack of uplink pilot signals received from different time slots. Specifically, the received signal is represented as a stacked matrix of different time slots on multiple subcarriers. Then, these stacked signals are correlated with each candidate angle and distance in the complete dictionary to generate a correlation matrix. This step is actually searching for the position that best matches the received signal with a candidate angle and distance in the complete dictionary.

[0121] When the full dictionary is used for peak search, the computational complexity of the MUSIC method is as high as O(M 3 )(where M is the number of antennas), which is particularly evident in THz near-field communication scenarios. At this time, the search based on the complete dictionary structure will lead to high computational complexity and storage requirements, and the performance of traditional algorithms will be greatly reduced in high dimensions. At the same time, due to the spherical wave characteristics and complex sparse structure of near-field signals, the MUSIC method significantly increases the amount of calculation and reduces the efficiency when performing eigenvalue decomposition and peak search. The ESPRIT method directly estimates θ by utilizing the rotational invariance of the signal subspace. AOA and θ AOD , also relies on the accuracy of the complete redundant dictionary to match the CSI, so the parameter estimation performance will also be greatly reduced in high-dimensional cases.

[0122] In MIMO systems in 6G terahertz scenarios, traditional dictionary designs usually cover a wide range of angular spaces, resulting in very large dictionary matrices. Such a complete dictionary is computationally very complex and requires a large amount of storage space. In practical applications, if a complete redundant dictionary is used in combination with the MUSIC or ESPRIT algorithm to extract CSI, it will not only lead to a degradation in parameter estimation performance, but also lead to low computational efficiency and waste of storage resources, especially in THz communication systems. This problem is particularly prominent.

[0123] In order to solve the above problems, this paper proposes a CSI extraction method based on dictionary optimization (Dictionary-Optimized CSI Extraction, DO-CSI). The CSI extraction method based on dictionary optimization is an efficient CSI extraction method. Through dictionary reduction and optimization, it effectively reduces the computational complexity and storage requirements. Compared with the traditional MUSIC and ESPRIT methods, DO-CSI can better adapt to the characteristics of complex near-field channels. Its main idea is to optimize the extraction of CSI information by gradually reducing the dictionary. The complete implementation process based on DO-CSI is as follows: Figure 2 As shown, the specific methods include:

[0124] Step 2.1: Receive RIS reflected signal y, where y = H RIS ·H BS s+n;

[0125] The H RIS represents the channel matrix of RIS, H BS represents the channel matrix of the base station, s represents the signal sent by the base station, and n represents the noise;

[0126] In the DO-CSI method, the antenna arrays at the transmitting and receiving ends are divided into multiple subarrays, each of which covers a specific directional angle. This division allows the system to process the received signal independently in each subarray. Specifically, the antenna arrays in the MIMO system are usually planned as multiple subarrays, each of which is connected by a specific RF link to ensure the independence of signal processing between subarrays. Under this structure, the wave propagation characteristics of the near-field channel can be modeled using a hybrid spherical-plane wave model, the signal propagation within the subarray can be processed using a plane wave model, and the spherical wave model is used between subarrays to capture the amplitude and phase changes of the signal.

[0127] Step 2.2: The first antenna subarray OMP is estimated to obtain the first sparse coefficient vector a1, and Θ ovs represents the initial complete redundant dictionary, and Ψ is the transformation basis of compressed sensing technology.

[0128] During the dictionary reduction process, a subarray (e.g., a subarray on the UE side) is first selected for preliminary channel estimation. By dividing the UE-side antenna array into multiple subarrays, the arrival angle and time delay of signals transmitted on different paths can be estimated separately. The independent processing of each subarray not only reduces the computational complexity, but also improves the accuracy of channel estimation, thereby further improving the accuracy of UE location. In addition, the division of subarrays can also help reduce the complexity of channel estimation, because each subarray can be estimated separately, and then the results can be combined to obtain a more accurate position estimate.

[0129] Compressed sensing is a signal processing technique applicable to recovering the original signal through a small number of linear measurements in cases where the signal is sparse or compressible. In the basic model of compressed sensing, it is assumed that the sparse signal x ∈ R N has a sparse representation under a certain transformation basis Ψ, that is, x = Ψs, where s is the sparse coefficient vector with only K non-zero elements (K << N). Through the measurement matrix Φ, the compressed observed value y ∈ R M is obtained, that is

[0130] y = Φx = ΦΨs

[0131] where Φ ∈ R M×N is the measurement matrix, satisfying M < N. The goal of compressed sensing is to reconstruct the sparse signal s through the observed value y, the known measurement matrix Φ, and the transformation basis Ψ. In the channel estimation problem, y represents the received signal, Θ = ΦΨ represents the dictionary matrix, and s represents the sparse channel response.

[0132] Step 2.3: Generate the support set supp(a1);

[0133] This step is usually completed by the OMP algorithm to obtain the initial support set supp(a1).

[0134] supp(a1): represents the support set of the sparse vector a1, that is, the set of indices containing non-zero elements. These indices correspond to specific columns in the dictionary matrix, indicating the positions of the sparse elements that best match the received signal.

[0135] Step 2.4: Based on the support set, optimize and reduce the initial complete redundant dictionary Θ ovs by retaining only the entries corresponding to the support set to form the optimized dictionary Θ DO , where Θ DO = Θ OVS [:, support(a1)];

[0136] Next, based on the estimation result of the first subarray, use this support set to optimize and reduce the initial complete redundant dictionary Θ ovs by retaining only the entries corresponding to the support set to form the optimized dictionary Θ DO . The specific formula is:

[0137] Θ DO = Θ ovs [:, supp(a1)] (4.18)

[0138] This optimization step significantly reduces the complexity of subsequent calculations because the estimation of subsequent subarrays only needs to be carried out in this reduced dictionary.

[0139] Step 2.5: Use the simultaneous orthogonal matching pursuit SOMP algorithm to estimate the channel and obtain the optimal solution of the channel matrix Extract channel state information CSI, and

[0140] To further improve the accuracy and efficiency of the estimation, the channel estimation of the subsequent subarrays uses the Simultaneous Orthogonal Matching Pursuit (SOMP) algorithm. The SOMP algorithm can consider the sparse structure of multiple subarrays at the same time, thus avoiding the information inconsistency problem that may occur when estimating subarrays one by one in the OMP method.

[0141] In the optimized dictionary, the SOMP algorithm gradually approaches the real CSI information by iteratively selecting the dictionary column most relevant to the current residual r. In each iteration, the dictionary column most relevant to the current residual is first calculated:

[0142]

[0143] Then, update the sparse representation coefficient α DO And the residual r:

[0144]

[0145] r=y-ΨΘ DO α DO (4.21)

[0146] The above calculation process is iterated continuously until the residual is small enough or the preset number of iterations is reached, and the sparse representation coefficient α is finally extracted. DO Corresponding to a specific θ AOA ,θ AOD and τ TOA Specifically, for θ AOA and θ AOD , assuming that the dictionary entry Θ DO The angle corresponding to [:,i] is θ i , then the extracted θ AOA and θ AOD They are:

[0147] θ AOA =θ Rx ,θ AOD =θ Tx (4.22)

[0148] Among them, θ Rx and θ Tx are the receiving and sending angles from RIS to UE respectively; for τ TOA , dictionary entry ΘDO [:,i] also contains the time delay information τ i , the extracted τ TOA for:

[0149] τ TOA =τ i (4.23)

[0150] The θ AOD represents the departure angle from RIS to UE, θ AOA represents the arrival angle from RIS to UE, τ TOA Indicates the arrival time from RIS to UE.

[0151] The DO-CSI method effectively reduces the computational complexity through dictionary reduction and optimization, and can accurately extract θ under high-dimensional and complex channel conditions. AOA ,θ AOD and τ TOA Compared with the traditional MUSIC and ESPRIT methods, DO-CSI performs better in THz near-field channels and can effectively cope with the challenges brought by high dimensionality and beam splitting effects in THz near-field MIMO channels.

[0152] Step 2.6: Output channel state information CS I and

[0153] Step 3: Design the reflection phase of the reconfigurable smart surface RIS and output the optimal beam direction;

[0154] like Figure 3 The complete RIS reflection phase design process is shown in Figure 3 As shown, specifically including:

[0155] Step 3.1: Obtain channel state information CSI (θ AOD ,θ AOA ,τ TOA );

[0156] Step 3.2: Generate initial codebook initial =f(θ1,θ2,…,θ k );

[0157] Furthermore, in step 3.2, the formula for generating the primary codebook is as follows:

[0158]

[0159] Among them, d n,m is the antenna spacing, λ is the signal wavelength, θ i It is the perspective between the target user and the RIS.

[0160] The primary codebook performs a rough beam scan with pre-set parameters and determines the preliminary beam direction in combination with the maximum received power strategy. Next, the phase of the primary codebook is further adjusted according to the propagation characteristics of the spherical wave so that it can better adapt to the beam changes caused by the spherical wave propagation process.

[0161] Step 3.3: Select the optimal initial beam direction θ opt , W represents a beamforming vector or a weight vector, which is used to control the transmission direction of the signal to maximize the received power; H represents the channel estimation matrix, which is the estimation result of the channel state information (CSI) and is used to describe the channel characteristics in the transmission path. In system design, by estimating the channel state information, the transmission strategy can be adjusted to improve the signal reception effect. r The channel matrix representing the reflection path describes the signal propagation path from the base station BS to the RIS and then to the user equipment UE. This matrix is ​​used to analyze and optimize the impact of the RIS on the signal reflection path. In discrete signal processing and beamforming, k represents the subcarrier index, which is used here to identify a specific subcarrier component. T represents the time index, which is used to describe the channel state that changes over time. In the RIS-assisted positioning system, dynamic adjustment of the system can be achieved through the channel state information at different times. Therefore represents the channel matrix generated by the Kth subcarrier at time node t. These parameters are combined to optimize the beamforming and channel estimation in the RIS-aided positioning system to achieve the best signal reception direction and maximize the received power.

[0162] Step 3.4: Generate high-level codebook high =phasecorrection(Codebook initial ), The φ corrected ( d ) represents the angle obtained after phase correction based on the initial codebook.

[0163] The advanced codebook is based on the primary codebook, and through refined beamforming, the system can better adapt to the propagation characteristics of spherical waves. The optimized CSI output is used for beamforming, real-time channel estimation adjustment and positioning.

[0164] In the RIS-assisted terahertz near-field positioning system, reflection phase design is a key step to achieve efficient beamforming and channel optimization. Beamforming in the terahertz band must not only consider the requirements of high sparsity and high resolution, but also deal with the challenges of the propagation characteristics of near-field spherical waves and the beam splitting effect. In order to better adapt to indoor positioning scenarios in terahertz near-field environments, the beam training part proposed in this application introduces a power distribution pattern (PDP) algorithm to supplement the existing beamforming strategy.

[0165] In terahertz near-field communication systems, due to the spherical wave characteristics of signal propagation, the design of beamforming needs to take into account the beam deviation effect. The beam deviation effect means that in spherical wave propagation, due to the non-planarity of the phase front, the beam may deviate from the expected direction, thus affecting the system performance. In order to overcome this effect, the beamforming strategy focuses on the following aspects:

[0166] (1) Beam focusing and spherical wave correction: In the terahertz near-field, beam forming not only relies on the plane wave assumption, but also must consider the influence of spherical wave propagation. By finely adjusting the phase of the beam, it can better adapt to the propagation characteristics of the spherical wave, thereby reducing the impact of beam deviation. In the case of spherical wave propagation, the phase change of the signal in space is not only linear, but also presents curvature. This curvature leads to the beam deviation effect, which is specifically manifested as the phase change of the spherical wave when the signal propagates between the RIS and the UE. It can be expressed as:

[0167]

[0168] Where d is the distance the signal propagates, R is the radius of curvature of the phase front in the beamforming process, and λ is the wavelength of the signal. In order to compensate for the spherical wave phase offset, a correction term is added to offset the effect so that the beam maintains good directivity in the target direction. The corrected beam phase can be expressed as:

[0169]

[0170] Among them, the correction This is to offset the additional phase shift caused by the spherical wave and ensure that the beam maintains the correct direction during propagation.

[0171] (2) Dynamic adjustment based on beam training: Combined with the real-time CSI feedback information during beam training, the reflection phase of RIS is dynamically adjusted to eliminate the beam offset caused by beam propagation. The traditional beam training strategy is based on the selection of maximum received power. In this process, the generation formula of the advanced codebook is as follows:

[0172] C s=(A H A) -1 A H G s (4.27)

[0173] Among them, C s is the advanced codebook, and A is the modified primary codebook matrix.

[0174] Step 3.5: Beam training based on power distribution pattern PDP;

[0175] In the broadband terahertz near-field positioning scenario, the RIS reflection phase strategy based on the hierarchical codebook design can select the appropriate beam direction through beam training, but it is easy to make mistakes in angle resolution. Especially in broadband signals, due to the frequency selectivity of the channel, confusion in beam direction selection may occur during beam training. In the traditional hierarchical codebook, the beam direction is often selected by the maximum received power in the process of refinement selection from primary beam to advanced beam. However, in terahertz near-field communication, due to the spherical wave characteristics of signal propagation and the beam splitting effect, the use of this method may lead to inaccurate direction selection, especially in the refinement beam selection stage, which is easy to confuse the beam selection results of different frequency bands, thereby affecting the positioning accuracy of the system. In order to address the above problems and further improve the accuracy of terahertz near-field angle estimation, a beam training strategy based on power distribution mode is introduced. The PDP algorithm provides an effective direction estimation optimization algorithm by analyzing the received power distribution of subcarriers in different directions. The derivation of the PDP strategy is based on antenna array theory. Specifically, the phase shift matrix of RIS can be expressed as:

[0176]

[0177] Among them, φ i represents the phase shift of RIS, and N is the number of reflection units of RIS. For each subcarrier, the power of the user received signal can be expressed as:

[0178]

[0179] Among them, h UE is the UE’s receive antenna array response vector, h RIS are the array response vectors of RIS. These vectors are based on a typical uniform linear array structure, where the antenna elements in the array are equally spaced and arranged in a straight line. The array response vector h RIS and h UE is obtained by accumulating phase differences, which are related to the position and incident angle (θ AOA / θ AOD ) related.

[0180] In the PDP strategy, the direction of the UE is analyzed and estimated by calculating the distribution pattern of the received power in space. Usually, two adjacent wide beams are set to capture the difference in power distribution received in space, so that the difference in received power in different directions can be calculated by the following formula:

[0181]

[0182] Where j1(φ) and j2(φ) represent the power responses of two adjacent beams received in the direction φ. Due to the beam splitting effect of subcarriers of different frequencies in a broadband system, the direction estimation in space may be offset, but by analyzing the power ratio ε(φ), the unique θ can be achieved. AOA and θ AOD more precise estimate of .

[0183] like Figure 4 As shown in the figure, the PDP algorithm sets a pair of adjacent wide beams by analyzing the received power distribution pattern, and calculates the power ratio within these beams. The beam training range in the figure is expressed as [μ-δ,μ+δ], where μ is the center angle and δ is the width of the training range. By analyzing the signal power received in the beam, the power ratio ε(φ) in each direction can be calculated, and the offset of the beam in that direction can be determined accordingly. Specifically, μ1=μ-δ and μ2=μ+δ shown in the figure represent the center angles of two groups of adjacent beams, respectively. By analyzing the power distribution of these beams, the directional offset caused by the beam splitting effect can be accurately estimated, and the directional estimation can be further optimized.

[0184] At the same time, the PDP-based algorithm also improves the beam training strategy, thereby enhancing the ability of RIS to reflect beams and locate UEs in real time, and can be better extended from the millimeter wave band to the wide beam positioning in the terahertz band.

[0185] Step 3.6: Confirm whether the current optimal beam direction is consistent with θ PDP Match, if not, return to step 3.3, otherwise go to step 3.7; the θ PDP is the UE beam direction calculated based on the power distribution pattern;

[0186] Step 3.7: Adjust the reconfigurable smart surface RIS reflection phase Φ so that the RIS reflection beam is accurately directed to the user equipment UE, and where ρ opt The phase parameter representing the optimized beam direction;

[0187] Step 3.8: Output the optimized channel state information CSI.

[0188] First, obtain CSI information to provide preliminary channel estimation for subsequent beam training. These CSI parameters are used to generate the initial codebook, which covers a relatively wide range of beam directions and is used to preliminarily lock the approximate position of the UE. The generation of the primary codebook is still based on the beam training principle of maximum received power, that is, the preliminary beam direction is determined by selecting the beam direction with the largest received power. On this basis, enter the advanced codebook generation stage. At this time, the beam direction is further optimized by combining the received power information fed back by the UE and the phase correction of the spherical wave. The generation of the advanced codebook not only needs to combine the feedback power information, but also needs to correct the spherical wave effect existing in the high-frequency band signal. Therefore, when generating the advanced codebook, the algorithm will adjust the phase of the beam in combination with the phase correction model to ensure that the signal can be more accurately focused on the area where the UE is located. The core of the generation of the advanced codebook is to refine it based on the preliminary CSI and the feedback information of the user equipment after the primary beam training. The CSI data fed back by the UE, including the received signal power information, is an important basis for determining the adjustment of the beam direction of the advanced codebook. In this process, the spherical wave effect will cause errors due to the phase change when the beam propagates in the high frequency band. Therefore, in the advanced codebook generation stage, the algorithm introduces a phase correction model to compensate for these errors. The phase correction of the spherical wave ensures that the beam can remain focused when propagating in the high frequency band, thereby improving the accuracy of the beam. At the same time, based on the generation of the second-level advanced codebook, the power distribution pattern (PDP) analysis is further introduced. The main task of PDP is to help further refine the beam direction by analyzing the received power distribution on different sub-beams. PDP can optimize the beam training results through power analysis of sub-beams with different frequencies. Especially in broadband systems, it can effectively solve the beam splitting and frequency selectivity effects. Therefore, after the advanced codebook is generated, the power distribution under the wide beam is calculated by PDP, which can reduce the angle range of the beam, so as to select the optimal beam direction within a smaller angle range. This process combines the information of power feedback and frequency selectivity to ensure that the direction of the beam can achieve the best performance on each subcarrier. Next, the angle consistency check is used to check whether the currently generated beam direction is consistent with the direction obtained by PDP analysis. If they are consistent, the next step is entered; if they are inconsistent, the algorithm returns to readjust the advanced codebook or beam direction until the two match. In this process, the algorithm checks whether the beam direction selected based on the maximum received power is consistent with the beam direction obtained by PDP analysis to ensure that the final beamforming is the result of optimization through the multi-layer feedback mechanism. If inconsistency is found during the check process, the system will readjust the phase or beam direction to ensure that each step can achieve the optimal beam pointing result.

[0189] In the RIS reflection phase adjustment, the phase of the RIS reflection unit is adjusted according to the previous angle consistency check results to ensure that the reflected beam is accurately aligned with the UE direction. The phase adjustment of the RIS unit combines the previous phase correction formula and further optimizes the beam pointing to ensure that the reflected beam can be effectively focused on the UE. Finally, the optimized CSI information is output, which contains the high-precision channel state information finally obtained after beam training and phase correction to provide support for subsequent communication and positioning. The entire process gradually refines the beam direction through hierarchical beam training, spherical wave correction, PDP analysis and feedback mechanism, and ensures that the RIS reflection phase design is optimized through UE feedback and power distribution mode evaluation in each step, and finally achieves high-precision channel estimation and positioning. The above RIS reflection phase strategy is not only used to improve the beam pointing accuracy, but also can improve the performance of compressed sensing channel estimation in real time through the optimized CSI information. The optimized CSI information can more accurately reflect the current channel status, allowing the OMP compressed sensing algorithm to better perform sparse representation and reconstruction of the channel matrix, further improving the accuracy of channel estimation, and finally outputting more optimized CSI information in real time for the correction of UE positioning information in the subsequent position estimation solution module, forming a two-way feedback module for channel estimation and beamforming optimization. This two-way optimization process ensures the adaptability and reliability of the system in complex communication environments, while also improving the overall robustness of the positioning system.

[0190] Step 4: The user equipment UE performs channel estimation again and outputs optimized channel state information CSI;

[0191] Step 5: Position solution and optimization based on maximum likelihood estimation (MLE).

[0192] Maximum Likelihood Estimation (MLE) is an algorithm for performing position estimation and calculating its positioning error bound PEB and orientation error bound OEB. MLE is used in this stage to optimize the user equipment position estimation after extracting the channel state information CSI. Its purpose is to find the position parameters that maximize the likelihood function related to the position, so as to accurately solve the optimal user equipment position estimation result.

[0193] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A reconfigurable smart surface-assisted terahertz indoor positioning algorithm based on dictionary optimization, characterized in that: include: Step 1: User equipment UE receives a signal randomly reflected by a reconfigurable smart surface RIS in the initial stage; Step 2: Channel estimation based on dictionary optimization and compressed sensing CS to obtain coarse estimated channel state information CSI; Step 3: Design the reflection phase of the reconfigurable smart surface RIS and output the optimal beam direction; Step 4: The user equipment UE performs channel estimation again and outputs optimized channel state information CSI; Step 5: Position solution and optimization based on maximum likelihood estimation (MLE).

2. The reconfigurable intelligent surface-assisted terahertz indoor positioning algorithm based on dictionary optimization according to claim 1 is characterized in that: Before step 1, the channel model of the reconfigurable intelligent surface-assisted terahertz indoor positioning system is established, which specifically includes determining the channel matrix H from the base station to the RIS. B,R , the channel matrix H from RIS to user equipment R,U and the phase control matrix Θ[n] of RIS on the nth subcarrier.

3. The reconfigurable intelligent surface-assisted terahertz indoor positioning algorithm based on dictionary optimization as claimed in claim 2 is characterized in that: The channel matrix H from the base station to the RIS B,R It is expressed as: Among them, N B is the number of base station antennas, N B is the number of reflection units in RIS, H B,R [n] is the channel matrix between the base station and the RIS on the nth subcarrier; N B and N R are the number of antennas of the base station and RIS respectively; L B,R is the number of multipaths between the base station and the RIS; α l,B,R [n] is the complex gain of the lth path at the nth subcarrier, which depends on the path loss and phase variation; and denote the response vectors of RIS and base station antenna array, respectively, which depend on the incident angle exit angle and path distance The channel matrix H from RIS to user equipment R,U , expressed as: Among them, H R,U [n] is the channel matrix between RIS and user equipment on the nth subcarrier; N U is the number of antennas on the user equipment; L R,U is the number of multipaths between RIS and user equipment; α l,R,U [n] is the complex gain of the lth path on the nth subcarrier, which depends on the path distance and environmental influences; and denote the response vectors of the user equipment and the RIS antenna array, respectively, which depend on the signal departure angle Angle of arrival and path distance 4. The reconfigurable intelligent surface-assisted terahertz indoor positioning algorithm based on dictionary optimization as claimed in claim 3 is characterized in that: The phase control matrix Θ[n] of the RIS on the nth subcarrier is expressed as: Where Θ[n] is the phase control matrix of RIS on the nth subcarrier, β i [n] is the reflection coefficient of the i-th RIS unit on the n-th subcarrier, θ i [n] is the phase offset of the i-th RIS unit.

5. The reconfigurable intelligent surface-assisted terahertz indoor positioning algorithm based on dictionary optimization as claimed in claim 4 is characterized in that: The specific method of obtaining the coarse estimated channel state information CSI based on the channel estimation of dictionary optimization and compressed sensing CS in step 2 includes: Step 2.1: Receive RIS reflected signal y, where y = H RIS ·H BS ·s+n, the H RIS represents the channel matrix of RIS, H BS represents the channel matrix of the base station, s represents the signal sent by the base station, and n represents the noise; Step 2.2: The first antenna subarray is orthogonally matched to pursue estimation, and the first sparse coefficient vector a1 is obtained, and Θ ovs represents the initial complete redundant dictionary, Ψ represents the transformation basis of the compressed sensing technology; Step 2.3: Generate support set supp(a1); Step 2.4: Based on the support set, the initial complete redundant dictionary Θ ovs Perform optimization and reduction, retaining only the entries corresponding to the support set, to form an optimized dictionary Θ DO , where Θ DO =Θ OVS [:,support(a1)]; Step 2.5: Use the simultaneous orthogonal matching pursuit SOMP algorithm to estimate the channel and obtain the optimal solution of the channel matrix Extract channel state information CSI, and Step 2.6: Output channel state information CSI and 6. The reconfigurable intelligent surface-assisted terahertz indoor positioning algorithm based on dictionary optimization as claimed in claim 5 is characterized in that: The simultaneous orthogonal matching pursuit SOMP algorithm specifically includes: (1) First calculate the dictionary column index i* that is most relevant to the current residual: Said T represents the transpose of the matrix; (2) Then, update the sparse representation coefficient α DO And the residual r: r=y-ΨΘ DO a DO (4.21) (3) The above process (1)-(2) is iterated continuously until the residual is small enough or the preset number of iterations is reached, and the sparse representation coefficient α is finally extracted. DO Corresponding to a specific θ AOA ,θ AOD and τ TOA Specifically, for θ AOA and θ AOD , assuming that the dictionary entry Θ DO The angle corresponding to [:,i] is θ i , then the extracted θ AOA and θ AOD They are: i AOA =θ Rx ,the AOD =θ Tx (4.22) Among them, θ Rx and θ Tx are the receiving and sending angles from RIS to UE respectively; for τ TOA , dictionary entry Θ DO [:,i] also contains the time delay information τ i , the extracted τ TOA for: t TOA =t i The θ AOD represents the departure angle from RIS to UE, θ AOA represents the arrival angle from RIS to UE, τ TOA Indicates the arrival time from RIS to UE.

7. The reconfigurable intelligent surface-assisted terahertz indoor positioning algorithm based on dictionary optimization according to claim 6 is characterized in that: The specific method of designing the reconfigurable smart surface RIS reflection phase and outputting the optimal beam direction in step 3 includes: Step 3.1: Obtain channel state information CSI (θ AOD ,θ AOA ,τ TOA ); Step 3.2: Generate initial codebook initial =f(θ1,θ2,…,θ k ); Step 3.3: Select the optimal initial beam direction θ opt , W represents the beamforming vector or weight vector, represents the channel estimation matrix, H r represents the channel matrix of the reflection path, k represents the subcarrier index, and t represents the time index, which is used to describe the channel state that changes over time. represents the channel matrix generated by the Kth subcarrier at time node t; Step 3.4: Generate high-level codebook high =phasecorrection(Codebook initial ), The φ corrected(d) It represents the angle obtained after phase correction based on the initial codebook. Step 3.5: Beam training based on power distribution pattern PDP; Step 3.6: Confirm whether the current optimal beam direction is consistent with θ PDP Match, if not, return to step 3.3, otherwise go to step 3.7; the θ PDP is the UE beam direction calculated based on the power distribution pattern; Step 3.7: Adjust the reconfigurable smart surface RIS reflection phase Φ so that the RIS reflection beam is accurately directed to the user equipment UE, and where ρ opt The phase parameter representing the optimized beam direction; Step 3.8: Output the optimized channel state information CSI.

8. The reconfigurable intelligent surface-assisted terahertz indoor positioning algorithm based on dictionary optimization according to claim 7 is characterized in that: In step 3.2, the formula for generating the primary codebook is as follows: Among them, d n,m is the antenna spacing, λ is the signal wavelength, θ i It is the perspective between the target user and the RIS.

9. The reconfigurable intelligent surface-assisted terahertz indoor positioning algorithm based on dictionary optimization according to claim 8, characterized in that: In step 3.4, the phase change of the spherical wave is Δφ s-wave Specifically: The corrected beam phase can be expressed as: Where d is the distance the signal travels, R is the radius of curvature of the phase front during beamforming, and λ is the wavelength of the signal. The generation formula of the advanced codebook is as follows: C s =(A H A) -1 A H G s (4.27) Among them, C s is the advanced codebook, and A is the modified primary codebook matrix.

10. The reconfigurable intelligent surface-assisted terahertz indoor positioning algorithm based on dictionary optimization according to claim 9, characterized in that: The specific method of beam training based on the power distribution mode PDP in step 3.5 is: The phase shift matrix of RIS can be expressed as: Among them, φ i represents the phase shift of RIS, N is the number of reflection units of RIS; for each subcarrier, the power of the user received signal can be expressed as: Among them, h UE is the UE’s receive antenna array response vector, h RIS are the array response vectors of the RIS; these vectors are based on a typical uniform linear array structure, where the antenna elements in the array are equally spaced and arranged in a straight line. RIS and h UE is obtained by accumulating phase differences, which are related to the position and incident angle (θ AOA / θ AOD ) related. In the PDP strategy, the direction of the UE is analyzed and estimated by calculating the distribution pattern of the received power in space. Usually, two adjacent wide beams are set to capture the difference in the power distribution received in space, so that the difference in received power in different directions is calculated by the following formula: Where j1(φ) and j2(φ) represent the power responses of two adjacent beams received in the direction φ. Due to the beam splitting effect of subcarriers of different frequencies in a broadband system, the direction estimation in space may be offset. However, by analyzing the power ratio ε(φ), the unique θ can be obtained. AOA and θ AOD more precise estimate of .

Images