Active node deployment and channel estimation method for IRS-assisted synaesthesia integrated system

Abstract

The invention discloses an IRS-assisted synaesthesia integration system active node deployment and channel estimation method, belonging to the field of wireless communication technology. The method includes: establishing a channel model based on the antenna array response and channel characteristics of the IRS-assisted synaesthesia integration system; setting the channel value of the channel model collected and observed by the active node on the IRS surface in the IRS-assisted synaesthesia integration system, and establishing the IRS receiving signal model after the base station sends a pilot to the IRS based on the channel value; establishing a channel graph topology structure matrix; cutting the channel graph topology structure matrix based on spectral clustering clustering, and selecting the smoothest node in each cluster for sampling, and deploying active nodes; introducing a graph smoothness constraint according to the local smoothness characteristics of the channel graph topology structure matrix, obtaining a linear objective function based on atomic norm minimization, solving to obtain a channel estimation result, and completing channel estimation. This method can ensure the accuracy and stability of channel estimation in different scenarios.

Description

Technical Field

[0001] The invention belongs to the technical field of wireless communications and relates to an IRS-assisted synaesthesia integrated system active node deployment and channel estimation method. Background Art

[0002] From the first generation of wireless communication (1G) that supports basic coverage to the large-scale and complex fifth generation mobile communication (5G) for large-scale commercial use, to the sixth generation mobile communication (6G) research work that is in full swing today, wireless communication technology continues to develop in depth. The new generation of mobile communication means higher transmission rate, lower latency, larger capacity, massive connections, etc. In this process, Integrated Sensing and Communication (ISAC) technology, as one of the core pillars of 6G, is gradually becoming a hot spot in the field of technology. No matter how rapidly wireless technology develops, channels have always been one of the most critical links in the process of achieving communication. Accurate identification and estimation of channels is the basis of information processing. Only based on accurate channel information can signal demodulation, interference elimination and power control be performed. Intelligent Reflection Surface (IRS), also known as Reconfigurable Intelligence Surface (RIS), is expected to become one of the key technologies of the new generation of wireless communication technology 6G and ISAC system with its advantages of low energy consumption, low cost and easy deployment. IRS consists of a large two-dimensional array of passive scattering elements. These elements control the electromagnetic properties of radio frequency waves so that reflected signals are coherently superimposed at the receiver or destructively reduce co-channel interference, which is beneficial to improving communication efficiency, coverage and overall network performance. Channel state information acquisition is one of the main challenges in the deployment of IRS-assisted systems. Existing channel estimation methods are divided into direct cascade channel estimation and separated cascade channel estimation. Typical direct cascade channel estimation methods include binary reflection method, minimization of Cramér-Rao lower bound method, deep learning-based convolutional neural network method, etc.; typical separated cascade channel estimation methods are divided into full passive estimation methods represented by iterative Bayesian inference algorithm, anchor-assisted two-stage channel estimation algorithm, parallel factor decomposition method according to active and passive estimation methods, and semi-passive estimation methods represented by using compressed sensing tools to construct channels at all IRS elements from channels seen only at active elements and implementing IRS and given channel interaction schemes based on deep learning.

[0003] However, in the face of the difficulty in obtaining channel state information due to the fact that systems based on active devices can directly obtain channel state information by sending training sequences, while IRS-assisted systems can only reflect signals but have no ability to actively send or receive, the existing RIS-assisted system channel estimation technology uses the method of deploying active sensors to estimate the channel state information of a large number of passive RIS with only a small number of training sequences, thereby saving pilot overhead to a great extent, but the semi-passive method still has some defects. Since this type of semi-passive estimation method is generally used for channels with strong correlation, and the hardware cost is increased when deploying active sensors, high-performance channel estimation of semi-passive IRS-assisted systems is still a major challenge. Graph signal processing provides a framework for analyzing signals defined in irregular domains or networks, making it suitable for IRS-assisted ISAC systems with complex signal environments. In recent years, related research technologies have considered the spatial correlation of antenna angles and amplitudes, and used graph signal methods and graph neural networks to reveal the application potential of graph signal processing in channel estimation. However, in current research and technology, there is still no work that directly uses graph signals for channel estimation without the help of graph neural networks. Summary of the invention

[0004] The purpose of the present invention is to solve the technical problem that the active node deployment and channel estimation method of the semi-passive IRS auxiliary system in the prior art has defects, and to provide an IRS-assisted synaesthesia integrated system active node deployment and channel estimation method.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] In a first aspect, the present invention discloses an IRS-assisted synaesthesia integrated system active node deployment and channel estimation method, comprising:

[0007] Establish an IRS-assisted synaesthesia integration system, and establish a channel model based on the antenna array response and channel characteristics of the IRS-assisted synaesthesia integration system;

[0008] An active node is set on the surface of an IRS in the IRS-assisted interaesthesia integrated system to collect and observe channel values ​​of the channel model, and an IRS receiving signal model is established after the base station sends a pilot to the IRS based on the channel values;

[0009] Establishing an objective function for channel estimation based on the IRS received signal model using a least squares method;

[0010] The objective function is optimized based on the smoothness of the graph signal and the characteristics of the Laplace matrix elements to obtain the optimized objective function under the graph Laplace energy constraint, and a channel graph topology structure matrix is ​​established based on the optimized objective function;

[0011] The channel graph topology matrix is ​​cut into clusters based on spectral clustering, and the smoothest node in each cluster is selected for sampling to deploy active nodes;

[0012] A graph smoothness constraint is introduced according to the local smoothness characteristics of the channel graph topology structure matrix, a linear objective function is obtained based on atomic norm minimization, a channel estimation result is obtained by solving, and channel estimation is completed.

[0013] Further improvements are:

[0014] The IRS-assisted synaesthesia integration system includes a multi-antenna base station BS, a multi-element intelligent reflection surface IRS and a multi-antenna mobile station MS; the antennas of the multi-antenna base station BS and the multi-antenna mobile station MS both adopt uniform linear arrays ULA; the multi-element intelligent reflection surface IRS adopts a uniform planar array UPA; the IRS surface is a multi-element intelligent reflection surface IRS; the base station is a multi-antenna base station BS.

[0015] The antenna array response of the IRS-assisted synaesthesia integration system specifically includes:

[0016] Assume that the channel directly connecting the multi-antenna base station BS and the multi-antenna mobile station MS in the IRS-assisted interawareness integrated system is blocked, and the multi-element intelligent reflective surface IRS is needed to assist the data transmission between the multi-antenna base station BS and the multi-antenna mobile station MS;

[0017] have The array response vector of the multi-antenna base station BS with root antennas is expressed as:

[0018]

[0019] in, represents the number of reflective elements of a multi-antenna base station BS; The wavelength is The wave number of Indicates wavelength; Indicates the antenna spacing; Indicates the departure angle; Represents an imaginary unit; symbol Represents the matrix Transpose operation; represents a natural constant;

[0020] have The receiving array response vector of a multi-element intelligent reflecting surface IRS is expressed as:

[0021]

[0022] in, Indicates the number of reflective elements of the multi-element intelligent reflective surface IRS; Indicates the number of elements in the horizontal direction of the array; Represents the number of elements in the vertical direction of the array; the size of the antenna array is expressed as ; Indicates the horizontal angle; Indicates the elevation angle; m Indicates the horizontal direction of the antenna array Column and ; n Indicates the vertical direction of the antenna array Line and ;

[0023] The multi-element intelligent reflecting surface IRS receiving array response is expressed as:

[0024]

[0025] in, represents the vertical response vector of the array; Represents the horizontal response vector of the array; symbol Represents the Kronecker product operation of two vectors; Represents the sine function operation; Represents the cosine function operation.

[0026] The channel model is established by the following steps:

[0027] The multi-element intelligent reflecting surface IRS is a semi-passive configuration, the propagation channel is a time division duplex mode and there is reciprocity, and bidirectional alternating downlink and uplink training is performed in the case of the semi-passive multi-element intelligent reflecting surface IRS;

[0028] The propagation channel includes two separate channels: the channel from the multi-antenna base station BS to the multi-element intelligent reflection surface IRS surface and the channel from the multi-element intelligent reflection surface IRS to the multi-antenna mobile station MS. Assuming that the propagation channel is a block fading channel, the propagation channel parameters remain unchanged within the coherence time; according to the geometric channel model, the propagation channel is defined as:

[0029]

[0030] Among them, the subscript represents the link propagation from a multi-antenna base station BS to a multi-element intelligent reflecting surface IRS; Represents the channel parameter matrix from the multi-antenna base station BS to the multi-element intelligent reflection surface IRS surface and belongs to dimensional imaginary field ; represents the number of resolvable paths; Indicates The path propagation gain of the paths; Indicates The horizontal angle of the path; Indicates The elevation angle of the path; Indicates the IRS receiving end The array response vector of the paths; Indicates The departure angle of each path; Indicates the BS sending end The conjugate transpose of the array response vector of the paths; the symbol It means taking the conjugate transpose of the matrix; express The departure angle vector composed of the departure angles of the paths; express The horizontal angle vector formed by the horizontal angles of the paths; express The elevation vector formed by the elevation angles of the paths; represents the array response matrix of the BS transmitter; represents the conjugate transpose of the array response matrix at the IRS receiver; express The path propagation gain matrix composed of the path propagation gains of the paths; operation It means constructing a vector into a matrix with the elements of the vector as the diagonal;

[0031] Considering the half-wavelength antenna unit spacing, the array response vectors of the BS transmitter and IRS receiver are expressed as:

[0032]

[0033] The array response matrix is ​​expressed as:

[0034]

[0035] in, represents the array response matrix of the IRS receiver, which is The IRS receiving end array response vectors of the paths are composed together; represents the array response matrix of the BS transmitter, which is composed of The BS transmitter array response vectors of the paths are composed together;

[0036] The channel from the multi-element intelligent reflecting surface IRS to the multi-antenna mobile station MS is expressed as:

[0037]

[0038] Among them, the subscript Represents the link propagation from the multi-element intelligent reflecting surface IRS to the multi-antenna mobile station MS; Represents the channel parameter matrix from the multi-element intelligent reflecting surface IRS surface to the multi-antenna mobile station MS and belongs to dimensional imaginary field ; is the number of antennas at the MS end; represents the number of resolvable paths; It is The propagation path gain of the path; Indicates The horizontal angle of the path; Indicates The elevation angle of the path; Indicates The angle of arrival of each path; and They are and Array response vector of ; express The arrival angle vector composed of the arrival angles of the paths; express The horizontal angle vector formed by the horizontal angles of the paths; express The elevation vector formed by the elevation angles of the paths; represents the array response matrix of the multi-antenna MS end; represents the conjugate transpose of the array response matrix of the multi-antenna IRS end; express A path propagation gain matrix composed of the path propagation gains of the paths;

[0039] The channel model from a multi-antenna base station BS to a multi-element intelligent reflection surface IRS surface, and then to a multi-antenna mobile station MS is expressed as:

[0040]

[0041] in, is a diagonal phase control matrix at the IRS with constant module entries on its diagonal, e.g., ,function Indicated by natural constant The exponential operation with base .

[0042] The IRS receiving signal model is established by the following steps:

[0043] Active nodes are arranged on the multi-element intelligent reflection surface IRS to collect observed channel values, and channel estimation is performed directly on the multi-element intelligent reflection surface IRS; two separate channels, a channel from a multi-antenna base station BS to the multi-element intelligent reflection surface IRS surface and a channel from the multi-element intelligent reflection surface IRS to a multi-antenna mobile station MS, are estimated separately;

[0044] The channel estimation from the multi-antenna base station BS to the multi-element intelligent reflection surface IRS surface is:

[0045] Assumptions IRS element blocks are placed An active node is used to process the signal. The signal model received by the multi-element intelligent reflecting surface IRS is expressed as:

[0046]

[0047] in, is the signal matrix received by IRS; is the sampling matrix; is additive Gaussian white noise; is the pilot sequence matrix.

[0048] The objective function in the objective function for channel estimation based on the IRS received signal model established by using the least squares method is specifically:

[0049]

[0050] Among them, the function Indicates seeking In the collection Minimum value on ; operation Indicates seeking The square of the norm.

[0051] The objective function is optimized based on the smoothness of the graph signal and the characteristics of the Laplace matrix elements to obtain the optimized objective function under the graph Laplace energy constraint, and the channel graph topology structure matrix is ​​established based on the optimized objective function, which specifically includes:

[0052] Based on the characteristics of channel propagation of IRS-assisted interaesthesia integrated system, each channel data of each element block from multi-antenna base station BS to multi-element intelligent reflective surface IRS is similar. Based on the local smoothness characteristics brought by this similarity, the graph signal is introduced to describe the channel state information.

[0053] Each element block of the multi-element intelligent reflective surface IRS is modeled as a vertex of an undirected weighted graph, and the corresponding channel data is a graph signal. ,in represents the set of all vertices, Represents the set of all edges, a symmetric non-negative matrix The weight matrix representing the undirected weighted graph;

[0054] The degree matrix of the undirected weighted graph is defined as:

[0055]

[0056] in, Indicates The degree of each node; is the weight matrix Position element, indicating The corresponding weight of the position;

[0057] The Laplacian matrix of the undirected weighted graph is defined as:

[0058]

[0059] Based on the graph signal formed by the IRS-assisted synaesthesia integration system, the simple position relationship between vertices is not enough to describe the correlation of channel information; the internal structure of the data is obtained from a large amount of data; since the channel information of adjacent element blocks should be similar, a structural matrix that can describe the internal structure of these data is obtained from multiple sets of prior channel data; define multiple sets of channel data , the smoothness of the channel data of the IRS-assisted synaesthesia integration system is measured by the quadratic form:

[0060]

[0061] in, is the vertex The real part of the channel complex signal; is the vertex The imaginary part of the channel complex signal; is the vertex The real part of the channel complex signal; is the vertex The imaginary part of the channel complex signal; is the vertex The weight of yes The data after sampling;

[0062] If strongly connected vertices have similar values, then the graph signal is smooth; the smaller the quadratic form of the above formula is, the smoother the signal on the graph is;

[0063] With the help of the smoothness of the graph signal, considering the energy constraint of the graph Laplacian and the element characteristics of the Laplacian matrix, the channel graph structure is obtained using the following optimization problem expression:

[0064]

[0065] in, It is observed The matrix constructed by group data; is the actual channel data matrix; operation Indicates finding the trace of a matrix; operation It means to find the square of Frobenius norm; and is the weight factor of the optimization problem; Representation Matrix No. elements; This term constrains the smoothness of the graph signal; Constrain the energy of the Laplacian graph; in order to avoid trivial solutions, introduce ; It constrains the elements of the Laplace matrix to ensure its non-negativity;

[0066] In the case of only given observations When grouping data, solve the following optimization problem to get :

[0067]

[0068] in, is the weight factor for this optimization problem.

[0069] The channel graph topology matrix is ​​cut into clusters based on spectral clustering, and the smoothest node in each cluster is selected for sampling, and the deployment of active nodes is specifically as follows:

[0070] The spectral clustering evolved from graph theory and is widely used in clustering. It structures data, regards data as nodes, and cuts the graph by comparing the edge weights between nodes. It divides the graph based on the normalized cut criterion, so that the ratio of the sum of the edge weights within each subgraph after the division and the sum of the edge weights between the subgraphs is minimized, thereby forming a clustering result that satisfies the maximum intra-cluster connection density and the minimum inter-cluster connection density.

[0071] Based on this, in order to obtain the cut graph of the undirected graph, the graph signal Cut into disconnected subgraphs, and the set of subgraphs is , the set of subgraphs satisfies , ; Define the weight between two subgraphs as:

[0072]

[0073] for subgraph points, define the graph cutting function:

[0074]

[0075] in, for The complement of

[0076] In order to avoid the phenomenon that the weight between subsets is the smallest but not the optimal cut in the minimum cut, and in order to reflect the vertex weights, the NCut spectral clustering cut method is considered:

[0077]

[0078] in, It is a cluster The “volume” of a cluster is the sum of all edge weights within the cluster.

[0079] for The synaesthesia integration system assisted by IRS element blocks is firstly spectral clustered considering the local smoothness of the channel data. For the spectral clustering of NCut, an indicator vector is introduced. ; For any indicator vector , it is a dimensional vector; also define:

[0080]

[0081] Based on this, calculation have to:

[0082]

[0083] For the indicator vector A subgraph of whose NCut optimization objective corresponds to ; then for The NCut expression of a subgraph is:

[0084]

[0085] in, for The matrix of indicator vectors, The NCut graph cutting of subgraphs is equivalent to minimizing , and because:

[0086]

[0087] That is to say , where the matrix is the unit matrix; then The optimal cut of the subgraph is expressed as the optimization objective:

[0088]

[0089] in, The parameter value that represents the minimum value of the objective function in its domain;

[0090] Each of the above indicator vectors is If is a matrix composed of unit orthogonal basis, is the standardized Laplace matrix, and the above optimization problem is directly transformed into finding the smallest characteristic values, that is , , , , the Laplace matrix is ​​normalized once;

[0091] Based on the graph Fourier transform, find the normalized Laplace matrix the smallest eigenvalues, and the corresponding feature vector, which The feature vectors form a Dimensional matrix; this matrix is ​​the Fourier basis matrix of the graph , the eigenvector corresponding to each eigenvalue is the indicator vector;

[0092] Using NCut graph cutting to perform spectral clustering to obtain cluster division The subgraph obtained after that cannot solve the sampling problem. It is necessary to select the nodes in each cluster and choose the one that best represents the The nodes with the subset features are used as sampling nodes;

[0093] To describe the overall smoothness of the graph, for the graph signal , define the total variation on the graph after a given Laplace matrix:

[0094]

[0095] in, Represents a vertex In the field of , the derivative is used to obtain the local total gradient on the graph:

[0096]

[0097] in, Indicates the overall change of a node in the graph signal, which is the difference value with other nodes; No. Rows are nodes The total gradient of all neighboring node signals; then define an expression to represent the difference between each node and other nodes:

[0098]

[0099] in, It is calculated A function of the mean of each row of the matrix; for The moment is defined as:

[0100]

[0101] in, Indicates the number of hops of the reachable path connecting the node to other nodes; No. The larger the element, the The more intense the change between this node and its neighbors within the hop, the lower the similarity;

[0102] Based on this, To represent the difference between nodes; calculate the local total gradient of each node in each cluster, and select the node corresponding to the smallest value as the sampling point of the subset. The value of is determined adaptively according to the cluster size and satisfies ,in is the total number of nodes in the current cluster; based on this, based on spectral clustering and considering the changes in the graph, we finally get a set of sampling points.

[0103] The step of introducing a graph smoothness constraint according to the local smoothness characteristics of the channel graph topology structure matrix, obtaining a linear objective function based on atomic norm minimization, and solving to obtain a channel estimation result, and completing the channel estimation specifically includes:

[0104] For the semi-passive IRS system, the structural information of the channel characterizes the internal connection of the channel data and is used to assist in channel estimation. Since the channel of the IRS system also has local smoothness, the optimization problem is obtained when performing channel estimation:

[0105]

[0106] And by In the two-dimensional Fourier dictionary, it has a sparse representation, so Rewritten as:

[0107]

[0108] in, Indicates The path propagation gain of the paths; and Represents the transmitting antenna array and receiving antenna array The spatial frequency of each path; represents the conjugate transpose of the transmit array response vector; represents the receiving end array response vector;

[0109] According to the spatial frequency belonging to 0-1, the atomic matrix is ​​defined , the resulting set of atoms is defined as:

[0110]

[0111] Then the atomic set The atomic norm expression of is:

[0112]

[0113] in, It means to find the infimum; yes The convex hull of

[0114] When the corresponding transmitter is a single-antenna scenario, it is a one-dimensional atomic norm minimization model; in a multi-antenna scenario, the corresponding semi-positive definite programming problem is:

[0115]

[0116] Among them, the transformation Indicates converting a vector into a Toeplitz matrix; is the channel vector corresponding to the single antenna case; is a vector describing the atomic structure; is a variable to be optimized;

[0117] Based on the smoothness of the graph and considering the sparse constraints of the atomic norm, the following convex problem is formed:

[0118]

[0119] in, is the regularization parameter; is the sampling matrix; is the pilot sequence; is the signal received at the active node; is the structural matrix of the constructed graph; atomic norm minimization is equivalent to a semi-positive definite problem, which can be further expressed as:

[0120]

[0121] in, is the target channel estimation result obtained after optimization.

[0122] In a second aspect, the present invention discloses an IRS-assisted synaesthesia integrated system active node deployment and channel estimation system, comprising:

[0123] A channel model building module is used to build an IRS-assisted synaesthesia integration system, and to build a channel model based on the antenna array response and channel characteristics of the IRS-assisted synaesthesia integration system;

[0124] An IRS receiving signal model establishment module, which is used to set a channel value of a channel model collected and observed by an active node on the surface of an IRS in the IRS-assisted interaesthesia integrated system, and to establish an IRS receiving signal model after the base station sends a pilot to the IRS based on the channel value;

[0125] An objective function establishment module, used to establish an objective function for channel estimation based on the IRS received signal model using a least squares method;

[0126] A channel graph topology structure matrix establishment module is used to optimize the objective function based on the smoothness of the graph signal and the characteristics of the Laplace matrix elements, obtain the optimized objective function under the graph Laplace energy constraint, and establish the channel graph topology structure matrix based on the optimized objective function;

[0127] An active node deployment module, used to cut the channel graph topology matrix into clusters based on spectral clustering, and select the smoothest node in each cluster for sampling to deploy active nodes;

[0128] The channel estimation module is used to introduce graph smoothness constraints according to the local smoothness characteristics of the channel graph topology structure matrix, obtain a linear objective function based on atomic norm minimization, solve and obtain a channel estimation result, and complete channel estimation.

[0129] Compared with the prior art, the present invention has the following beneficial effects:

[0130] The present invention discloses an IRS-assisted synaesthesia integrated system active node deployment and channel estimation method, which optimizes the objective function by introducing the smoothness of graph signals and the characteristics of Laplace matrix elements, constructs the optimization objective function under the graph Laplace energy constraint, and establishes the channel graph topology matrix. This processing method can capture the channel characteristics more accurately, effectively improving the accuracy of channel estimation. At the same time, by spectral clustering and clustering graph cutting and selecting the smoothest node for sampling, the optimized deployment of active nodes is achieved, and the efficiency of channel estimation is further improved. Compared with the traditional method, the present method utilizes the IRS-assisted synaesthesia integrated system, collects the observed channel values ​​through active nodes, and establishes a receiving signal model based on these channel values ​​for channel estimation. This method can significantly reduce the number of required training sequences, thereby greatly reducing the pilot overhead and improving the resource utilization efficiency of the system. The present method breaks through the limitations of the prior art and directly uses graph signals for channel estimation for the first time without the aid of graph neural networks. By introducing graph smoothness constraints and atomic norm minimization, a linear objective function is obtained, and the channel estimation result is solved. This innovation not only simplifies the complexity of channel estimation, but also provides new ideas and directions for future channel estimation technology. The channel model and received signal model established by this method can be applied to different types of IRS-assisted interaural integration systems, with good flexibility and adaptability. At the same time, by optimizing the objective function and the construction of the channel graph topology matrix, this method can cope with changes in channel characteristics in different scenarios and ensure the accuracy and stability of channel estimation. BRIEF DESCRIPTION OF THE DRAWINGS

[0131] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention 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 creative work.

[0132] Figure 1 It is a flow chart of the active node deployment and channel estimation method of the IRS-assisted synaesthesia integration system in the present invention;

[0133] Figure 2 A model diagram of the IRS-assisted synaesthesia integration system of the present invention;

[0134] Figure 3 It is a schematic diagram comparing the optimal cut graph and the minimum cut graph in the present invention;

[0135] FIG4 (a) is a performance comparison diagram of mean square error of channel estimation using OMP recovery under different sampling methods as the signal-to-noise ratio varies in an embodiment of the present invention;

[0136] FIG4 (b) is a performance comparison diagram of the mean square error of channel estimation using OMP recovery under different sampling methods in an embodiment of the present invention as the number of active nodes changes;

[0137] FIG4 (c) is a performance comparison diagram of mean square error of channel estimation based on graph smoothing recovery with signal-to-noise ratio under different sampling methods in an embodiment of the present invention;

[0138] FIG4 (d) is a performance comparison diagram of the mean square error of channel estimation based on graph smoothing recovery under different sampling methods in an embodiment of the present invention as the number of active nodes changes;

[0139] FIG5( a ) is a sampling pattern using uniform sampling in an embodiment of the present invention;

[0140] FIG5( b ) is a sampling pattern of minimum local total gradient sampling based on spectral clustering in an embodiment of the present invention;

[0141] FIG6 (a) is a performance comparison diagram of mean square error of channel estimation with signal-to-noise ratio variation using the improved atomic norm minimization recovery method based on graph smoothing constraints and the atomic norm minimization recovery method under different sampling methods in an embodiment of the present invention;

[0142] FIG6 (b) is a performance comparison diagram of the mean square error of the minimum local total gradient sampling based on spectral clustering of different channel estimation methods in an embodiment of the present invention as the signal-to-noise ratio changes;

[0143] FIG6 (c) is a performance comparison diagram of the mean square error of the minimum local total gradient sampling based on spectral clustering of different channel estimation methods in an embodiment of the present invention as the number of active nodes changes;

[0144] Figure 7 It is a module diagram of the active node deployment and channel estimation system of the IRS-assisted synaesthesia integration system in the present invention. DETAILED DESCRIPTION

[0145] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0146] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0147] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0148] Terminology explanation:

[0149] 1. 1G: First generation mobile communication system.

[0150] 2. 5G: Fifth generation mobile communication system.

[0151] 3. 6G: Sixth generation mobile communication system.

[0152] 4. ISAC: Integrated Sensing and Communication.

[0153] 5. MIMO: Multiple Inputs Multiple Outputs.

[0154] 6. IRS: Intelligent Reflection Surface.

[0155] 7. OMP: Orthogonal Matching Pursuit.

[0156] 8. BS: multi-antenna base station, Base Station.

[0157] 9. MS: multi-antenna mobile station, Mobile Station.

[0158] 10. ULA: Uniform Line Array.

[0159] 11. UPA: Uniform Plane Array.

[0160] 12. Image Signal GRefers to an undirected graph. In mathematics and graph theory, an undirected graph is a structure consisting of a set of vertices (or nodes) and a set of undirected edges between these vertices. In an undirected graph, edges have no direction, that is, if there is an edge between vertex A and vertex B, then we can say that A is connected to B, or we can say that B is connected to A. In this description of spectral clustering, the graph G is used to represent the data structure, in which data points are regarded as nodes (vertices) of the graph, and the similarity (or distance) between nodes is represented by the weight of the edge. The goal of spectral clustering is to achieve clustering by cutting the graph (i.e. splitting the graph into several subgraphs) so that the connection weights between nodes in the same subgraph (or cluster) are as high as possible, while the connection weights between different subgraphs are as low as possible. Therefore, the graph signal G Here it is an abstract representation used to model the relationship between data points and is segmented by the spectral clustering algorithm to achieve clustering.

[0161] The present invention is further described in detail below in conjunction with the accompanying drawings:

[0162] See also Figure 1 The embodiment of the present invention discloses an IRS-assisted synaesthesia integrated system active node deployment and channel estimation method, including:

[0163] S1, establish an IRS-assisted synaesthesia integration system, and establish a channel model based on the antenna array response and channel characteristics of the IRS-assisted synaesthesia integration system;

[0164] S2, setting an active node on the surface of an IRS in the IRS-assisted interaesthesia integrated system to collect channel values ​​of an observed channel model, and establishing an IRS receiving signal model after the base station sends a pilot to the IRS based on the channel values;

[0165] S3, establishing an objective function for channel estimation based on the IRS received signal model using a least squares method;

[0166] S4, optimizing the objective function based on the smoothness of the graph signal and the characteristics of the Laplace matrix elements to obtain an optimized objective function under the graph Laplace energy constraint, and establishing a channel graph topology matrix based on the optimized objective function;

[0167] S5, cutting the channel graph topology structure matrix into clusters based on spectral clustering, selecting the smoothest node in each cluster for sampling, and deploying active nodes;

[0168] S6, introducing a graph smoothness constraint according to the local smoothness characteristics of the channel graph topology structure matrix, obtaining a linear objective function based on atomic norm minimization, solving to obtain a channel estimation result, and completing the channel estimation.

[0169] The present invention is based on a semi-passive IRS-assisted synaesthesia integrated system, and proposes a solution based on graph signal processing for the IRS active node location deployment problem and channel estimation problem. First, the problem of active node deployment is converted into a graph sampling problem, and a graph signal is modeled based on channel data. The structural matrix of the channel is obtained for the data, and signal data processing based on graph signals rather than graph neural networks is realized. Then, using the structural information, clustering is performed based on spectral clustering, and after obtaining the clustering of the optimal cut graph, sampling points are selected for the nodes in the cluster, and the smoothest node is determined to represent the cluster. A sampling method with the minimum local total gradient based on spectral clustering is further formed, which effectively selects key nodes and realizes a simpler and more efficient semi-passive node deployment. Then, considering the local smoothness of the data, the constraint of graph smoothing is introduced to restrict the channel, and an improved atomic norm minimization method based on graph smoothing is formed to perform channel estimation, and the internal structure information of the data is effectively used by using the local smoothness characteristics of the graph to realize high-performance channel estimation. Finally, the sampling method and channel estimation method of the present invention effectively improve the system performance.

[0170] The content of the present invention is described in detail below in conjunction with specific embodiments:

[0171] Step 1: Construct an IRS-assisted ISAC system and establish a communication channel model based on the antenna array response and channel characteristics.

[0172] Specifically, Figure 2 As shown in FIG. 1 , this embodiment considers constructing an IRS-assisted multiple input multiple output (MIMO) communication and target detection perception scenario, namely, an integrated sensing and communication (ISAC) system. The system consists of a multi-antenna base station (BS), a multi-element intelligent reflecting surface (IRS), a multi-antenna mobile station (MS), and a point target. For the communication task, the number of antennas on the BS and MS and the number of IRS reflective elements are respectively , and The system assumes that the channel directly connecting the BS to the MS is blocked, and the IRS is needed to assist the data transmission between the BS and the MS. The antennas of the BS and the MS are both uniform line arrays (ULA), and the surface of the IRS is uniform plane array (UPA).

[0173] Based on this, The BS-side array response vector of the root antenna can be expressed as:

[0174]

[0175] in, represents the number of reflective elements of a multi-antenna base station BS; The wavelength is The wave number of Indicates the antenna spacing; Indicates the departure angle; Represents an imaginary unit; symbol Represents the matrix Transpose operation; represents a natural constant;

[0176] For those with The receive array response vector can be expressed as:

[0177]

[0178] in, Indicates the number of reflective elements of the multi-element intelligent reflective surface IRS; Indicates the number of elements in the horizontal direction of the array; Represents the number of elements in the vertical direction of the array; the size of the antenna array is expressed as ; Indicates the horizontal angle; Indicates the elevation angle; m Indicates the horizontal direction of the antenna array Column and ; n Indicates the vertical direction of the antenna array Line and ;

[0179] Furthermore, the IRS receiving array response can be expressed as:

[0180]

[0181]

[0182]

[0183] in, represents the vertical response vector of the array; Represents the horizontal response vector of the array; symbol Represents the Kronecker product of two vectors; Represents the sine function operation; Represents the cosine function operation.

[0184] Downlink data training and transmission are considered in a semi-passive IRS configuration. The channel is in time division duplex mode and reciprocity exists. In the case of a semi-passive IRS, downlink and uplink training are performed in both directions. The propagation channel consists of two separate channels: the channel from the BS to the IRS surface and the channel from the IRS to the MS. and Assuming that the channel is a block fading channel, the channel parameters remain unchanged during the coherence time. According to the geometric channel model, the channel can be defined as:

[0185]

[0186] Among them, the subscript represents the link propagation from a multi-antenna base station BS to a multi-element intelligent reflecting surface IRS; Represents the channel parameter matrix from the multi-antenna base station BS to the multi-element intelligent reflection surface IRS surface and belongs to dimensional imaginary field ; represents the number of resolvable paths; Indicates The path propagation gain of the paths; Indicates The horizontal angle of the path; Indicates The elevation angle of the path; Indicates the IRS receiving end The array response vector of the paths; Indicates The departure angle of each path; Indicates the BS sending end The conjugate transpose of the array response vector of the paths; the symbol It means taking the conjugate transpose of the matrix; express The departure angle vector composed of the departure angles of the paths; express The horizontal angle vector formed by the horizontal angles of the paths; express The elevation vector formed by the elevation angles of the paths; represents the array response matrix of the BS transmitter; represents the conjugate transpose of the array response matrix at the IRS receiver; express The path propagation gain matrix composed of the path propagation gains of the paths; operation It means constructing a vector into a matrix with the elements of the vector as the diagonal;

[0187] Considering half-wavelength antenna element spacing, the array response vector is expressed as:

[0188]

[0189]

[0190] Based on this, the array response matrix is ​​defined as:

[0191]

[0192]

[0193] in, represents the array response matrix of the IRS receiver, which is The IRS receiving end array response vectors of the paths are composed together; represents the array response matrix of the BS transmitter, which is composed of The BS transmitter array response vectors of the paths are composed together;

[0194] Similarly, the channel IRS-MS from the multi-element intelligent reflecting surface IRS to the multi-antenna mobile station MS is expressed as:

[0195]

[0196] Among them, the subscript Represents the link propagation from the multi-element intelligent reflecting surface IRS to the multi-antenna mobile station MS; Represents the channel parameter matrix from the multi-element intelligent reflecting surface IRS surface to the multi-antenna mobile station MS and belongs to dimensional imaginary field ; is the number of antennas at the MS end; represents the number of resolvable paths; It is The propagation path gain of the path; Indicates The horizontal angle of the path; Indicates The elevation angle of the path; Indicates The angle of arrival of each path; and They are and Array response vector of ; express The arrival angle vector composed of the arrival angles of the paths; express The horizontal angle vector formed by the horizontal angles of the paths; express The elevation vector formed by the elevation angles of the paths; represents the array response matrix of the multi-antenna MS end; represents the conjugate transpose of the array response matrix of the multi-antenna IRS end; express The path propagation gain matrix is ​​composed of the path propagation gain of the paths; based on this, the channel from the entire base station to the IRS and then to the mobile station can be expressed as:

[0197]

[0198] in, is a diagonal phase control matrix at the IRS with constant module entries on its diagonal, e.g., ,function Indicated by natural constant The exponential operation with base .

[0199] Step 2: Active nodes are arranged on the IRS surface to collect observed channel values ​​and construct an IRS receiving signal model after the base station sends a pilot signal to the IRS.

[0200] Specifically, in order to reduce pilot overhead, active nodes are arranged on the IRS surface to collect observed channel values. At this time, channel estimation can be performed directly on the IRS surface. After the active sensors are arranged, the BS-IRS and IRS-MS channels can be estimated separately. Since the two channel estimation methods are similar, only the channel estimation at the BS-IRS end is discussed and used as an example.

[0201] Assume that K active sensors that can process signals are placed in N IRS element blocks. In this hybrid IRS system, the received signal is no longer sent by the base station to the IRS, reflected by the IRS surface, coupled to the IRS-MS channel, and then the channel is estimated at the terminal. In fact, when estimating the downlink channel, the base station sends a pilot to the IRS and then directly estimates the channel on the IRS surface. Therefore, the signal received by the IRS is expressed as:

[0202]

[0203] in, is the signal matrix received by IRS; is the sampling matrix; is additive Gaussian white noise; is the pilot sequence matrix.

[0204] Step 3: Establish an optimization objective function by minimizing the error between the transmitted signal and the received signal according to the least squares criterion.

[0205] Specifically, this embodiment aims to realize RIS-assisted semi-passive MIMO system channel estimation, and uses the least squares criterion to establish the objective function of channel estimation after the active node receives the signal:

[0206]

[0207] Among them, the function Indicates seeking In the collection Minimum value on ; operation Indicates seeking The square of the norm.

[0208] Step 4: Using the smoothness of the graph signal and the characteristics of the Laplace matrix elements, establish the optimization objective function under the graph Laplace energy constraint and construct the channel graph topology.

[0209] Specifically, in order to deploy active nodes, the problem of sensor node placement is equivalent to a sampling problem. Considering the characteristics of channel propagation of the IRS-assisted system, each channel data of each element block from the BS to the IRS should be similar. Based on the local smoothness characteristics brought by this similarity, the graph signal method is introduced to describe the channel state information.

[0210] Preferably, each IRS element block is modeled as a graph vertex, and the corresponding channel data is the graph signal. ,in represents the set of all vertices, Represents the set of all edges, a symmetric non-negative matrix Represents the weight matrix of this undirected weighted graph.

[0211] The degree matrix of this graph is defined as:

[0212]

[0213] in, Indicates The degree of each node; is the weight matrix Position element, indicating The corresponding weight of the position;

[0214] And the Laplacian matrix of this graph is defined as:

[0215]

[0216] Based on the graph signal formed by IRS auxiliary system modeling, the simple position relationship between vertices is not enough to describe the correlation of channel information. The internal structure of the data can be obtained from a large amount of data. Since the channel information of adjacent element blocks should be similar, a structural matrix that can describe the internal structure of these data can be obtained from multiple sets of prior channel data. Define multiple sets of channel data , the smoothness of the channel data of the IRS auxiliary system can be measured by the quadratic form:

[0217]

[0218] in, is the vertex The real part of the channel complex signal; is the vertex The imaginary part of the channel complex signal; is the vertex The real part of the channel complex signal; is the vertex The imaginary part of the channel complex signal; is the vertex The weight of It is subsampled data;

[0219] With the help of the smoothness of the graph signal, considering the energy constraint of the graph Laplacian and the element characteristics of the Laplacian matrix, the channel graph structure is obtained using the following optimization problem expression:

[0220]

[0221]

[0222] in, It is observed The matrix constructed by group data; is the actual channel data matrix; operation Indicates finding the trace of a matrix; operation It means to find the square of Frobenius norm; and is the weight factor of the optimization problem; Representation Matrix No. elements; This term constrains the smoothness of the graph signal; Constrain the energy of the Laplacian graph; in order to avoid trivial solutions, introduce ; It constrains the elements of the Laplace matrix to ensure its non-negativity;

[0223] Preferably, only given observed When grouping data, the following optimization problem can be solved to obtain :

[0224]

[0225]

[0226] in, is the weight factor for this optimization problem.

[0227] Step 5: After obtaining the channel structure matrix from the data, the graph is cut into clusters based on spectral clustering, and the smoothest node in each cluster is selected for sampling to achieve the deployment of active nodes.

[0228] Specifically, spectral clustering evolved from graph theory and is widely used in clustering. It structures data, regards data as nodes, and cuts the graph by comparing the edge weights between nodes. Based on the normalized cut criterion, the graph is partitioned to minimize the ratio of the sum of the edge weights within each subgraph after the partition to the sum of the edge weights between the subgraphs, thereby forming a clustering result that satisfies the maximum intra-cluster connection density and the minimum inter-cluster connection density.

[0229] Based on this, in order to obtain the cut graph of the undirected graph, the graph signal Cut into disconnected subgraphs, and the set of subgraphs is , the set of subgraphs satisfies , . Define the weight between two subgraphs as:

[0230]

[0231] for subgraph points, define the graph cutting function:

[0232]

[0233] in, for The complement of .

[0234] Further, such as Figure 3 As shown in the figure, in order to avoid the phenomenon that the weight between subsets is the smallest but not the optimal cut, and in order to reflect the vertex weight, the NCut spectral clustering cut method is considered:

[0235]

[0236] in, It is a cluster The “volume” of a cluster is the sum of all edge weights within the cluster.

[0237] Preferably, for data with internal structure, the sampling pattern should not be designed by simply performing random sampling or uniform sampling. The sampling pattern should be designed according to the internal structure and channel characteristics. The placement of active nodes in the IRS channel is similar. In order to improve the performance of channel estimation as much as possible, the active nodes should be placed at relatively "critical" positions on the IRS surface.

[0238] Specifically, for Considering the local smoothness of the channel data, the spectral clustering is first performed for the system assisted by IRS element blocks. For the spectral clustering of NCut, an indicator vector is first introduced For any indicator vector , it is a dimensional vector. Also define:

[0239]

[0240] Based on this, calculation have to:

[0241]

[0242] Furthermore, for the indicator vector A subgraph of whose NCut optimization objective corresponds to . Then for The NCut expression of a subgraph is:

[0243]

[0244] in, is the trace of the matrix. That is to say, at this time The NCut graph cutting of subgraphs is equivalent to minimizing . And because:

[0245]

[0246] That is to say .but The optimal cut of a subgraph can be expressed as the optimization objective:

[0247]

[0248] in, The parameter value that represents the minimum value of the objective function in its domain;

[0249] Each indicator vector in the above optimization problem is N-dimensional, which is an NP-hard problem. is a matrix composed of unit orthogonal basis, is the standardized Laplace matrix, and the above optimization problem is directly transformed into finding the smallest eigenvalues. ,in is the matrix composed of eigenvectors; , , It is to standardize the Laplace matrix.

[0250] Furthermore, based on the graph Fourier transform, we find the standardized the smallest The corresponding eigenvalues ​​can be obtained feature vector, which The feature vectors form a Dimensional matrix. This matrix is ​​the Fourier basis matrix of the graph , the eigenvector corresponding to each eigenvalue is the indicator vector. Since the dimensional constraint is approximated when solving the optimization problem, some information will be lost. The matrix cannot fully indicate the belonging of each sample. Each row of the matrix is ​​clustered once using traditional clustering, such as K-means.

[0251] Using NCut graph cutting to perform spectral clustering to obtain cluster division The subgraph obtained after that cannot solve the sampling problem. Furthermore, it is necessary to select the nodes in each cluster and select a node that best represents the characteristics of the subset as the sampling node.

[0252] Specifically, to describe the overall smoothness of the graph, for the graph signal , define the total variation on the graph after a given Laplace matrix:

[0253]

[0254] in, Represents a vertex The neighborhood of . Taking the derivative gives the local total gradient on the graph:

[0255]

[0256] in, The sum of a row represents the overall change of a node in the graph signal, that is, the difference value with other nodes. No. Rows are nodes The total gradient of all node signals adjacent to it. So we can define an expression to represent the difference between each node and other nodes:

[0257]

[0258] in, It is calculated A function of the mean of each row of the matrix. The moment is defined as:

[0259]

[0260] in, Indicates the number of hops of the reachable path connecting the node to other nodes; No. The larger the element, the The more intense the change between this node and its neighbors within a hop, the lower the similarity.

[0261] Based on this, as shown in Table 1, it is the minimum local total gradient sampling algorithm based on spectral clustering. To represent the difference between nodes. Calculate the local total gradient of each node in each cluster, and select the node corresponding to the smallest value as the sampling point of the subset. The value of is determined adaptively according to the cluster size and satisfies ,in is the total number of nodes in the current cluster. Based on this, based on spectral clustering and considering the changes in the graph, a set of sampling points can be finally obtained.

[0262] Table 1 Minimum local gradient sampling algorithm based on spectral clustering;

[0263]

[0264] Step 6: Introduce graph smoothness constraints based on the local smoothness characteristics of the data, obtain the linear objective function based on atomic norm minimization, and solve to obtain the channel estimation result.

[0265] Specifically, for the semi-passive IRS system, the structural information of the channel represents the internal connection of the channel data. Such internal structural relationship can assist in channel estimation. Since the channel of the IRS system should also have local smoothness, when performing channel estimation, the optimization problem can be obtained:

[0266]

[0267] And by In the two-dimensional Fourier dictionary, it has a sparse representation, so Rewritten as:

[0268]

[0269] in, Indicates The path propagation gain of the paths; and Represents the transmitting antenna array and receiving antenna array The spatial frequency of each path; represents the conjugate transpose of the transmit array response vector; represents the receiving end array response vector;

[0270] According to the spatial frequency belonging to 0-1, the atomic matrix is ​​defined , the resulting set of atoms is defined as:

[0271]

[0272] Then the atomic set The atomic norm expression of is:

[0273]

[0274] in, It means to find the infimum; yes The convex hull of .

[0275] Preferably, this embodiment considers the scenario where the corresponding transmitting end is a single antenna, that is, a one-dimensional atomic norm minimization model. It can still be generalized in a multi-antenna scenario. The corresponding semi-positive definite programming problem is:

[0276]

[0277] Among them, the transformation Indicates converting a vector into a Toeplitz matrix. is the channel vector corresponding to the single antenna case, is a vector describing the atomic structure, is a variable to be optimized;

[0278] Finally, based on the smoothness of the graph and considering the sparse constraints of the atomic norm, the following convex problem is formed:

[0279]

[0280] in: is the regularization parameter; is the sampling matrix; is the pilot sequence; is the signal received at the active node; is the structural matrix of the constructed graph. Atomic norm minimization is equivalent to a semi-positive definite problem, which can be further expressed as:

[0281]

[0282] in, is the target channel estimation result obtained after optimization.

[0283] The above optimization problem is a semi-definite programming problem with a linear objective function, which is solvable in polynomial time. At this time, the traditional CVX toolkit can be used to solve the problem directly.

[0284] Example 1

[0285] For example, this embodiment sets a The intelligent reflective surface auxiliary system with 100 element blocks is simulated. The number of channel multipaths is equal to , base station transmitting antenna The number is equal to 1, and the IRS arranges several sensors that can receive data, and randomly generates a set of arrival angles and departure angles. This embodiment assumes that the channel is time reciprocal and assumes that multiple sets of channel data are known a priori. The simulation is set to have 200 sets of channel data a priori. This embodiment processes the two hundred sets of data, obtains the Laplace matrix, and then starts the experimental simulation. Since the base station and IRS are usually fixed in position in the actual system, this embodiment considers this section of the channel to be quasi-static. During the simulation, a set of departure angles and arrival angles are randomly generated, and this angle is used to first generate 200 sets of channel data with different channel gains as the channel data for acquiring the structure matrix. Then, the same angle information is used to change the channel gain to generate 2000 sets of channel data for simulation experiments.

[0286] This embodiment first simulates the proposed sampling method to compare the performance of different sampling methods in channel estimation using a simple channel estimation method, with the sampling rate set to 1 / 4. , considering that there are 4 points in each group when the average grouping is performed, set the variable representing the number of hops in the algorithm . As shown in Table 2, this embodiment uses A-opt sampling, E-opt sampling, D-opt sampling, uniform sampling and topology-aware graph sampling as comparison methods for the proposed sampling schemes in the simulation. Taking into account the layout cost of the actual system, random sampling is usually not considered when comparing performance, but this embodiment still gives the results in the simulation. At the same time, as shown in Table 3, this embodiment adopts the channel estimation methods used in semi-passive channel estimation for channel estimation, namely OMP (Orthogonal Matching Pursuit), graph-based smoothing recovery and atomic norm minimization (ANM).

[0287] Table 2 Comparison of sampling methods

[0288]

[0289] Table 3 Comparison of recovery methods

[0290]

[0291] Referring to Figure 4 (a) and Figure 4 (b), the sparse characteristics of the channel are utilized, OMP is used as the recovery method, and the algorithm performance of random sampling, uniform sampling, E-opt, topology-aware graph sampling, and the minimum local total gradient sampling method based on spectral clustering of the present invention are simulated and compared. It can be seen that the sampling method of the present invention is significantly better than the results of other sampling methods when OMP is used for channel estimation, indicating that the selection of sampling positions will indeed improve the channel estimation performance. When the number of active nodes is less than 23, the sampling method of the present invention is the best; when it is greater than 23, the method of the present invention is only second to the method based on graph topology perception proposed in the literature, but the difference is not very large. Therefore, the proposed sampling method can achieve good results when using the OMP method for channel estimation.

[0292] Referring to Figure 4 (c) and Figure 4 (d), the channel estimation is performed based on the graph smoothing method, and the simulation compares the performance of uniform sampling, random sampling, A-opt, E-opt, D-opt, topology-aware graph sampling, and the minimum local total gradient sampling method based on spectral clustering of the present invention. It can be clearly seen that the sampling method of the present invention has a greatly improved performance compared with other sampling methods when performing channel estimation under the restriction of graph smoothing. At the same time, regardless of the number of active nodes, the sampling scheme of the present invention can achieve the best recovery performance.

[0293] See Figure 5 (a) and Figure 5 (b), which are the sampling pattern of this embodiment and the uniformly distributed sampling pattern. It can be clearly seen that the sampling distribution of the sampling method of the present invention is very uneven, and it is not a discrete distribution, and the sampling is relatively clustered. In horizontal and vertical comparison, this sampling scheme captures more vertical nodes, and the distribution within a row is relatively sparse. At the same time, the edge nodes are not selected, so that the clustering information can correspond. By capturing one node in each cluster and the sampling rate reaches 1 / 4, the size of each set obtained will not be too large. The more clustered the graph changes in relatively dense places, the more sampling points are needed. However, since what is needed is the most distinctive node in each partition, it is difficult for the edge block to have such a role, so it is generally not captured.

[0294] See Figure 6 (a), which is the simulation result of the improved atomic norm minimization method based on graph smoothness constraint of this embodiment. The method is represented by GANM in the figure and compared with the ANM method. As can be seen from the figure, the sampling method of the present invention has a certain performance improvement compared with the atomic norm minimization. Especially when the active nodes are evenly distributed and based on the distribution of E-opt. It shows that the atomic norm minimization method based on graph smoothness constraint is effective. At the same time, it shows that the channel state information of the IRS auxiliary system is internally structured information, and the performance of channel estimation can be improved through this local smoothness starting from the structure. Not only that, from the change of the sampling method in the figure, the performance is greatly improved when the proposed sampling is used for channel estimation, which once again illustrates the performance superiority of the sampling method of the present invention.

[0295] Referring to FIG. 6 (b) and FIG. 6 (c), based on the sampling method of the present invention, the performance comparison of OMP, the recovery method of the present invention, ANM, and the recovery method based on graph smoothing is simulated respectively. As shown in the figure, OMP has the worst channel estimation performance. Under low signal-to-noise ratio conditions, the performance of channel estimation based on graph smoothing is the best. However, at medium and high signal-to-noise ratios, the atomic norm minimization channel estimation method based on graph smoothing constraints of the present invention has the best performance, which improves the performance of channel estimation. In addition, the channel estimation performance improves with the increase in the number of active nodes, and the performance of atomic norm minimization and improved atomic norm minimization based on smoothness constraints both improves with the increase in active nodes, indicating that the method of the present invention has the best performance.

[0296] See also Figure 7 The present invention discloses an IRS-assisted synaesthesia integrated system active node deployment and channel estimation system, comprising:

[0297] A channel model building module is used to build an IRS-assisted synaesthesia integration system, and to build a channel model based on the antenna array response and channel characteristics of the IRS-assisted synaesthesia integration system;

[0298] An IRS receiving signal model establishment module, which is used to set a channel value of a channel model collected and observed by an active node on the surface of an IRS in the IRS-assisted interaesthesia integrated system, and to establish an IRS receiving signal model after the base station sends a pilot to the IRS based on the channel value;

[0299] An objective function establishment module, used to establish an objective function for channel estimation based on the IRS received signal model using a least squares method;

[0300] A channel graph topology structure matrix establishment module is used to optimize the objective function based on the smoothness of the graph signal and the characteristics of the Laplace matrix elements, obtain the optimized objective function under the graph Laplace energy constraint, and establish the channel graph topology structure matrix based on the optimized objective function;

[0301] An active node deployment module, used to cut the channel graph topology matrix into clusters based on spectral clustering, and select the smoothest node in each cluster for sampling to deploy active nodes;

[0302] The channel estimation module is used to introduce graph smoothness constraints according to the local smoothness characteristics of the channel graph topology structure matrix, obtain a linear objective function based on atomic norm minimization, solve and obtain a channel estimation result, and complete channel estimation.

[0303] The third object of the present invention is to provide an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the steps of the active node deployment and channel estimation method of the IRS-assisted synaesthesia integrated system are implemented.

[0304] A fourth object of the present invention is to provide a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the IRS-assisted synaesthesia integrated system active node deployment and channel estimation method are implemented.

[0305] The IRS-assisted synaesthesia integrated system active node deployment and channel estimation method comprises the following steps:

[0306] Establish an IRS-assisted synaesthesia integration system, and establish a channel model based on the antenna array response and channel characteristics of the IRS-assisted synaesthesia integration system;

[0307] An active node is set on the surface of an IRS in the IRS-assisted interaesthesia integrated system to collect channel values ​​of an observed channel model, and an IRS receiving signal model is established after the base station sends a pilot to the IRS based on the channel values;

[0308] Establishing an objective function for channel estimation based on the IRS received signal model using a least squares method;

[0309] The objective function is optimized based on the smoothness of the graph signal and the characteristics of the Laplace matrix elements to obtain the optimized objective function under the graph Laplace energy constraint, and a channel graph topology structure matrix is ​​established based on the optimized objective function;

[0310] The channel graph topology matrix is ​​cut into clusters based on spectral clustering, and the smoothest node in each cluster is selected for sampling to deploy active nodes;

[0311] A graph smoothness constraint is introduced according to the local smoothness characteristics of the channel graph topology structure matrix, a linear objective function is obtained based on atomic norm minimization, a channel estimation result is obtained by solving, and channel estimation is completed.

[0312] It should be understood by those skilled in the art that the embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM (Compact Disc Read-Only Memory), optical storage, etc.) containing computer-usable program code.

[0313] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0314] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0315] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0316] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An IRS-assisted synaesthesia integrated system active node deployment and channel estimation method, characterized in that: include: Establish an IRS-assisted synaesthesia integration system, and establish a channel model based on the antenna array response and channel characteristics of the IRS-assisted synaesthesia integration system; An active node is set on the surface of an IRS in the IRS-assisted interaesthesia integrated system to collect and observe channel values ​​of the channel model, and an IRS receiving signal model is established after the base station sends a pilot to the IRS based on the channel values; Establishing an objective function for channel estimation based on the IRS received signal model using a least squares method; The objective function is optimized based on the smoothness of the graph signal and the characteristics of the Laplace matrix elements to obtain the optimized objective function under the graph Laplace energy constraint, and a channel graph topology structure matrix is ​​established based on the optimized objective function; The objective function is optimized based on the smoothness of the graph signal and the characteristics of the Laplace matrix elements to obtain the optimized objective function under the graph Laplace energy constraint, and the channel graph topology structure matrix is ​​established based on the optimized objective function, which specifically includes: Based on the characteristics of channel propagation of IRS-assisted interaesthesia integrated system, each channel data of each element block from multi-antenna base station BS to multi-element intelligent reflective surface IRS is similar. Based on the local smoothness characteristics brought by this similarity, the graph signal is introduced to describe the channel state information. Each element block of the multi-element intelligent reflective surface IRS is modeled as a vertex of an undirected weighted graph, and the corresponding channel data is a graph signal. ,in represents the set of all vertices, Represents the set of all edges, a symmetric non-negative matrix The weight matrix representing the undirected weighted graph; The degree matrix of the undirected weighted graph is defined as: in, Indicates The degree of each node; is the weight matrix Position element, indicating The corresponding weight of the position; Indicates the number of reflective elements of the multi-element intelligent reflective surface IRS; The Laplacian matrix of the undirected weighted graph is defined as: Based on the graph signal formed by the IRS-assisted synaesthesia integration system, the simple position relationship between vertices is not enough to describe the correlation of channel information; the internal structure of the data is obtained from a large amount of data; since the channel information of adjacent element blocks should be similar, a structural matrix that can describe the internal structure of these data is obtained from multiple sets of prior channel data; define multiple sets of channel data , the smoothness of the channel data of the IRS-assisted synaesthesia integration system is measured by the quadratic form: in, is the vertex The real part of the channel complex signal; is the vertex The imaginary part of the channel complex signal; is the vertex The real part of the channel complex signal; is the vertex The imaginary part of the channel complex signal; is the vertex The weight of yes The data after sampling; If strongly connected vertices have similar values, then the graph signal is smooth; the smaller the quadratic form of the above formula is, the smoother the signal on the graph is; With the help of the smoothness of the graph signal, considering the energy constraint of the graph Laplacian and the element characteristics of the Laplacian matrix, the channel graph structure is obtained using the following optimization problem expression: in, It is observed The matrix constructed by group data; is the actual channel data matrix; operation Indicates finding the trace of a matrix; operation It means to find the square of Frobenius norm; and is the weight factor of the optimization problem; Representation Matrix No. elements; This term constrains the smoothness of the graph signal; Constrain the energy of the Laplacian graph; in order to avoid trivial solutions, introduce ; It constrains the elements of the Laplace matrix to ensure its non-negativity; In the case of only given observations When grouping data, solve the following optimization problem to get : in, is the weight factor of the optimization problem; The channel graph topology matrix is ​​cut into clusters based on spectral clustering, and the smoothest node in each cluster is selected for sampling to deploy active nodes; A graph smoothness constraint is introduced according to the local smoothness characteristics of the channel graph topology structure matrix, a linear objective function is obtained based on atomic norm minimization, a channel estimation result is obtained by solving, and channel estimation is completed.

2. The IRS-assisted synaesthesia integrated system active node deployment and channel estimation method according to claim 1, characterized in that: The IRS-assisted synaesthesia integration system includes a multi-antenna base station BS, a multi-element intelligent reflection surface IRS and a multi-antenna mobile station MS; the antennas of the multi-antenna base station BS and the multi-antenna mobile station MS both adopt uniform linear arrays ULA; the multi-element intelligent reflection surface IRS adopts a uniform planar array UPA; the IRS surface is a multi-element intelligent reflection surface IRS; the base station is a multi-antenna base station BS.

3. The IRS-assisted synaesthesia integrated system active node deployment and channel estimation method according to claim 2, characterized in that: The antenna array response of the IRS-assisted synaesthesia integration system specifically includes: Assume that the channel directly connecting the multi-antenna base station BS and the multi-antenna mobile station MS in the IRS-assisted interawareness integrated system is blocked, and the multi-element intelligent reflective surface IRS is needed to assist the data transmission between the multi-antenna base station BS and the multi-antenna mobile station MS; have The array response vector of the multi-antenna base station BS with root antennas is expressed as: in, represents the number of reflective elements of a multi-antenna base station BS; The wavelength is The wave number of Indicates wavelength; Indicates the antenna spacing; Indicates the departure angle; Represents an imaginary unit; symbol Represents the matrix Transpose operation; represents a natural constant; have The receiving array response vector of a multi-element intelligent reflecting surface IRS is expressed as: in, Indicates the number of reflective elements of the multi-element intelligent reflective surface IRS; Indicates the number of elements in the horizontal direction of the array; Represents the number of elements in the vertical direction of the array; the size of the antenna array is expressed as ; Indicates the horizontal angle; Indicates the elevation angle; m Indicates the horizontal direction of the antenna array Column and ; n Indicates the vertical direction of the antenna array Line and ; The multi-element intelligent reflecting surface IRS receiving array response is expressed as: in, represents the vertical response vector of the array; Represents the horizontal response vector of the array; symbol Represents the Kronecker product operation of two vectors; Represents the sine function operation; Represents the cosine function operation.

4. The IRS-assisted synaesthesia integrated system active node deployment and channel estimation method according to claim 3, characterized in that: The channel model is established by the following steps: The multi-element intelligent reflecting surface IRS is a semi-passive configuration, the propagation channel is a time division duplex mode and there is reciprocity, and bidirectional alternating downlink and uplink training is performed in the case of the semi-passive multi-element intelligent reflecting surface IRS; The propagation channel includes two separate channels, a channel from a multi-antenna base station BS to a multi-element intelligent reflecting surface IRS surface and a channel from a multi-element intelligent reflecting surface IRS to a multi-antenna mobile station MS. Assuming that the propagation channel is a block fading channel, the propagation channel parameters remain unchanged within the coherence time. According to the geometric channel model, the propagation channel is defined as: Among them, the subscript represents the link propagation from a multi-antenna base station BS to a multi-element intelligent reflecting surface IRS; Represents the channel parameter matrix from the multi-antenna base station BS to the multi-element intelligent reflection surface IRS surface and belongs to dimensional imaginary field ; represents the number of resolvable paths; Indicates The path propagation gain of the paths; Indicates The horizontal angle of the path; Indicates The elevation angle of the path; Indicates the IRS receiving end The array response vector of the paths; Indicates The departure angle of each path; Indicates the BS sending end The conjugate transpose of the array response vector of the paths; the symbol It means taking the conjugate transpose of the matrix; express The departure angle vector composed of the departure angles of the paths; express The horizontal angle vector formed by the horizontal angles of the paths; express The elevation vector formed by the elevation angles of the paths; represents the array response matrix of the BS transmitter; represents the conjugate transpose of the array response matrix at the IRS receiver; express The path propagation gain matrix composed of the path propagation gains of the paths; operation It means constructing a vector into a matrix with the elements of the vector as the diagonal; Considering the half-wavelength antenna unit spacing, the array response vectors of the BS transmitter and IRS receiver are expressed as: The array response matrix is ​​expressed as: in, represents the array response matrix of the IRS receiver, which is The IRS receiving end array response vectors of the paths are composed together; represents the array response matrix of the BS transmitter, which is composed of The BS transmitter array response vectors of the paths are composed together; The channel from the multi-element intelligent reflecting surface IRS to the multi-antenna mobile station MS is expressed as: Among them, the subscript Represents the link propagation from the multi-element intelligent reflecting surface IRS to the multi-antenna mobile station MS; Represents the channel parameter matrix from the multi-element intelligent reflecting surface IRS surface to the multi-antenna mobile station MS and belongs to dimensional imaginary field ; is the number of antennas at the MS end; represents the number of resolvable paths; It is The propagation path gain of the path; Indicates The horizontal angle of the path; Indicates The elevation angle of the path; Indicates The angle of arrival of each path; and They are and Array response vector of ; express The arrival angle vector composed of the arrival angles of the paths; express The horizontal angle vector formed by the horizontal angles of the paths; express The elevation vector formed by the elevation angles of the paths; represents the array response matrix of the multi-antenna MS end; represents the conjugate transpose of the array response matrix of the multi-antenna IRS end; express A path propagation gain matrix composed of the path propagation gains of the paths; The channel model from a multi-antenna base station BS to a multi-element intelligent reflection surface IRS surface, and then to a multi-antenna mobile station MS is expressed as: in, is the diagonal phase control matrix at the IRS with constant module entries on its diagonal, ,function Indicated by natural constant The exponential operation with base .

5. The IRS-assisted synaesthesia integrated system active node deployment and channel estimation method according to claim 4, characterized in that: The IRS receiving signal model is established by the following steps: Active nodes are arranged on the multi-element intelligent reflecting surface IRS to collect observed channel values, and channel estimation is performed directly on the multi-element intelligent reflecting surface IRS; Two separate channels, a channel from a multi-antenna base station BS to a multi-element intelligent reflecting surface IRS and a channel from a multi-element intelligent reflecting surface IRS to a multi-antenna mobile station MS, are estimated separately; The channel estimation from the multi-antenna base station BS to the multi-element intelligent reflection surface IRS surface is: Assumptions IRS element blocks are placed An active node is used to process the signal. The signal model received by the multi-element intelligent reflecting surface IRS is expressed as: in, is the signal matrix received by IRS; is the sampling matrix; is additive Gaussian white noise; is the pilot sequence matrix.

6. The IRS-assisted synaesthesia integrated system active node deployment and channel estimation method according to claim 5, characterized in that: The objective function in the objective function for channel estimation based on the IRS received signal model established by using the least squares method is specifically: Among them, the function Indicates seeking In the collection Minimum value on ; operation Indicates seeking The square of the norm.

7. The IRS-assisted synaesthesia integrated system active node deployment and channel estimation method according to claim 6, characterized in that: The channel graph topology matrix is ​​cut into clusters based on spectral clustering, and the smoothest node in each cluster is selected for sampling, and the deployment of active nodes is specifically as follows: The spectral clustering evolved from graph theory and is widely used in clustering. It structures data, regards data as nodes, and cuts the graph by comparing the edge weights between nodes. It divides the graph based on the normalized cut criterion, so that the ratio of the sum of the edge weights within each subgraph after the division and the sum of the edge weights between the subgraphs is minimized, thereby forming a clustering result that satisfies the maximum intra-cluster connection density and the minimum inter-cluster connection density. Based on this, in order to obtain the cut graph of the undirected graph, the graph signal Cut into disconnected subgraphs, and the set of subgraphs is , the set of subgraphs satisfies , ; Define the weight between two subgraphs as: for subgraph points, define the graph cutting function: in, for The complement of In order to avoid the phenomenon that the weight between subsets is the smallest but not the optimal cut in the minimum cut, and in order to reflect the vertex weights, the NCut spectral clustering cut method is considered: in, It is a cluster The "volume" of a cluster is the sum of all edge weights within the cluster. for The synaesthesia integration system assisted by IRS element blocks is firstly spectral clustered considering the local smoothness of the channel data. For the spectral clustering of NCut, an indicator vector is introduced. ; For any indicator vector , it is a dimensional vector; also define: Based on this, calculation have to: in, represents the set of all vertices, Represents a set of vertices The vertices; For the indicator vector A subgraph of whose NCut optimization objective corresponds to ; then for The NCut expression of a subgraph is: in, for The matrix of indicator vectors, The NCut graph cutting of subgraphs is equivalent to minimizing , and because: That is to say , where the matrix is the unit matrix; then The optimal cut of the subgraph is expressed as the optimization objective: in, The parameter value that represents the minimum value of the objective function in its domain; Each of the above indicator vectors is If is a matrix composed of unit orthogonal basis, is the standardized Laplace matrix, and the above optimization problem is directly transformed into finding the smallest characteristic values, that is , , , , the Laplace matrix is ​​normalized once; Based on the graph Fourier transform, find the normalized Laplace matrix the smallest eigenvalues, and the corresponding feature vector, which The feature vectors form a Dimensional matrix; this matrix is ​​the Fourier basis matrix of the graph , the eigenvector corresponding to each eigenvalue is the indicator vector; Using NCut graph cutting to perform spectral clustering to obtain cluster division The subgraph obtained after that cannot solve the sampling problem. It is necessary to select the nodes in each cluster and choose the one that best represents the The nodes with the subset features are used as sampling nodes; To describe the overall smoothness of the graph, for the graph signal , define the total variation on the graph after a given Laplace matrix: in, Represents a vertex In the field of , the derivative is used to obtain the local total gradient on the graph: in, Indicates the overall change of a node in the graph signal, which is the difference value with other nodes; No. Rows are nodes The total gradient of all neighboring node signals; then define an expression to represent the difference between each node and other nodes: in, It is calculated A function of the mean of each row of the matrix; for The moment is defined as: in, Indicates the number of hops of the reachable path connecting the node to other nodes; No. The larger the element, the The more intense the change between this node and its neighbors within the hop, the lower the similarity; Based on this, To represent the difference between nodes; calculate the local total gradient of each node in each cluster, and select the node corresponding to the smallest value as the sampling point of the subset. The value of is determined adaptively according to the cluster size and satisfies ,in is the total number of nodes in the current cluster; based on this, based on spectral clustering and considering the changes in the graph, we finally get a set of sampling points.

8. The IRS-assisted synaesthesia integrated system active node deployment and channel estimation method according to claim 7, characterized in that: The step of introducing a graph smoothness constraint according to the local smoothness characteristics of the channel graph topology structure matrix, obtaining a linear objective function based on atomic norm minimization, and solving to obtain a channel estimation result, and completing the channel estimation specifically includes: For the semi-passive IRS system, the structural information of the channel characterizes the internal connection of the channel data and is used to assist in channel estimation. Since the channel of the IRS system also has local smoothness, the optimization problem is obtained when performing channel estimation: And by In the two-dimensional Fourier dictionary, it has a sparse representation, so Rewritten as: in, Indicates The path propagation gain of the paths; and Represents the transmitting antenna array and receiving antenna array The spatial frequency of each path; represents the conjugate transpose of the transmit array response vector; represents the receiving end array response vector; According to the spatial frequency belonging to 0-1, the atomic matrix is ​​defined , the resulting set of atoms is defined as: Then the atomic set The atomic norm expression of is: in, It means to find the infimum; yes The convex hull of When the corresponding transmitter is a single-antenna scenario, it is a one-dimensional atomic norm minimization model; in a multi-antenna scenario, the corresponding semi-positive definite programming problem is: Among them, the transformation Indicates converting a vector into a Toeplitz matrix; is the channel vector corresponding to the single antenna case; is a vector describing the atomic structure; is a variable to be optimized; Based on the smoothness of the graph and considering the sparse constraints of the atomic norm, the following convex problem is formed: in, is the regularization parameter; is the sampling matrix; is the pilot sequence; is the signal received at the active node; is the structural matrix of the constructed graph; atomic norm minimization is equivalent to a semi-positive definite problem, expressed as: in, is the target channel estimation result obtained after optimization.

9. An IRS-assisted synaesthesia integrated system active node deployment and channel estimation system based on the method of claim 1, characterized in that: include: A channel model building module is used to build an IRS-assisted synaesthesia integration system, and to build a channel model based on the antenna array response and channel characteristics of the IRS-assisted synaesthesia integration system; An IRS receiving signal model establishment module, which is used to set a channel value of a channel model collected and observed by an active node on the surface of an IRS in the IRS-assisted interaesthesia integrated system, and to establish an IRS receiving signal model after the base station sends a pilot to the IRS based on the channel value; An objective function establishment module, used to establish an objective function for channel estimation based on the IRS received signal model using a least squares method; A channel graph topology structure matrix establishment module is used to optimize the objective function based on the smoothness of the graph signal and the characteristics of the Laplace matrix elements, obtain the optimized objective function under the graph Laplace energy constraint, and establish the channel graph topology structure matrix based on the optimized objective function; An active node deployment module is used to cut the channel graph topology matrix into clusters based on spectral clustering, and select the smoothest node in each cluster for sampling to deploy active nodes; The channel estimation module is used to introduce graph smoothness constraints according to the local smoothness characteristics of the channel graph topology structure matrix, obtain a linear objective function based on atomic norm minimization, solve and obtain a channel estimation result, and complete channel estimation.

Images