[go: up one dir, main page]

CN117939402A - CSI indoor positioning method and device based on graph neural network - Google Patents

CSI indoor positioning method and device based on graph neural network Download PDF

Info

Publication number
CN117939402A
CN117939402A CN202311830099.3A CN202311830099A CN117939402A CN 117939402 A CN117939402 A CN 117939402A CN 202311830099 A CN202311830099 A CN 202311830099A CN 117939402 A CN117939402 A CN 117939402A
Authority
CN
China
Prior art keywords
csi
features
graph
indoor positioning
sub
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202311830099.3A
Other languages
Chinese (zh)
Inventor
余官定
叶子琦
肖棋琦
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Jinhua Research Institute Of Zhejiang University
Original Assignee
Jinhua Research Institute Of Zhejiang University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Jinhua Research Institute Of Zhejiang University filed Critical Jinhua Research Institute Of Zhejiang University
Priority to CN202311830099.3A priority Critical patent/CN117939402A/en
Publication of CN117939402A publication Critical patent/CN117939402A/en
Pending legal-status Critical Current

Links

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
    • H04W4/02—Services making use of location information
    • H04W4/029—Location-based management or tracking services
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W16/00—Network planning, e.g. coverage or traffic planning tools; Network deployment, e.g. resource partitioning or cells structures
    • H04W16/22—Traffic simulation tools or models
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
    • H04W4/30—Services specially adapted for particular environments, situations or purposes
    • H04W4/33—Services specially adapted for particular environments, situations or purposes for indoor environments, e.g. buildings

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Position Fixing By Use Of Radio Waves (AREA)

Abstract

本发明公开了一种基于图神经网络的CSI室内定位方法及装置，属于室内定位技术领域，包括：在室内定位场景中布置若干个接入点，在定位区域内设置若干个参考点；获取用户终端在不同参考点处与若干个接入点之间的CSI指纹特征，并从中提取幅度特征和相位特征作为节点特征，根据接入点之间的拓扑信息得到边特征和邻接矩阵，完成图结构的构建，进而得到图数据集；使用图数据集离线训练CSI‑GNN，直至损失函数最小化或达到设定的迭代次数；将训练好的CSI‑GNN算法模型用于在线定位，输出用户终端的实时位置。本发明提出的CSI‑GNN算法能够动态评估并优化幅度信息和相位信息的相对重要性，从而提升室内定位的准确性和稳定性。

The present invention discloses a CSI indoor positioning method and device based on graph neural network, which belongs to the field of indoor positioning technology, including: arranging several access points in the indoor positioning scene, setting several reference points in the positioning area; obtaining CSI fingerprint features between the user terminal and several access points at different reference points, and extracting amplitude features and phase features as node features, obtaining edge features and adjacency matrix according to the topological information between the access points, completing the construction of the graph structure, and then obtaining a graph data set; using the graph data set to train CSI-GNN offline until the loss function is minimized or the set number of iterations is reached; using the trained CSI-GNN algorithm model for online positioning, and outputting the real-time position of the user terminal. The CSI-GNN algorithm proposed in the present invention can dynamically evaluate and optimize the relative importance of amplitude information and phase information, thereby improving the accuracy and stability of indoor positioning.

Description

CSI indoor positioning method and device based on graph neural network
Technical Field
The invention belongs to the technical field of indoor positioning, and particularly relates to a CSI indoor positioning method and device based on a graph neural network.
Background
In the field of WiFi indoor positioning, with the wide deployment of wireless networks and the popularization of intelligent devices, various WiFi-based positioning methods have become hot spots for research and application. The Received Signal Strength Indicator (RSSI) technique is one of the most common methods in WiFi indoor positioning. It estimates the location of the user equipment based on the signal strength received from different access points. But in a complex indoor environment, its positioning accuracy is significantly reduced by physical obstructions and multipath effects.
The indoor positioning technology of channel state Information (CHANNEL STATE Information) is a more advanced wireless positioning technology, and can better cope with multipath effect and environmental change, thereby realizing higher positioning precision. The CSI contains detailed propagation information of the radio signal on each subcarrier, including amplitude attenuation and phase offset. This information reflects multipath effects such as reflections, diffractions and scattering of the signal when it encounters an obstacle. Since CSI can capture subtle signal changes, it can be used to sense environmental changes such as: the user's location in the indoor environment changes.
A common indoor positioning method based on CSI is to create a CSI fingerprint map, i.e. to collect CSI data at different locations in the indoor environment. And during positioning, matching the CSI data acquired in real time with the data in the fingerprint database to determine the current position.
Patent document with publication number CN110109055A discloses an indoor positioning method based on RSSI ranging, comprising: the method comprises the steps of adopting a square topological structure to carry out substation node arrangement, utilizing ZigBee technology networking to carry out communication, carrying out data broadcasting by a tag node, sending a tag data packet, analyzing by the substation node, then forming the substation data packet, sending the substation data packet to a base station node, extracting RSSI values in the data packet by the base station node, calculating the distance between the tag node and each substation node, and carrying out repeated optimization on the distance values through Kalman filtering and an iteration method to eliminate the influence of external interference on the accuracy of the distance values, thereby ensuring the positioning accuracy. However, the RSSI adopted in the present invention is greatly affected by signal attenuation and multipath effects, which may cause instability of signal strength, so that the positioning accuracy is limited, and in addition, there are many factors that may interfere with signals in the indoor environment, such as: walls, furniture, etc., which also affect the propagation and reception of signals, and thus the accuracy of the RSSI positioning.
Patent document with publication number CN111212379a discloses a novel CSI indoor positioning method based on convolutional neural network, comprising: obtaining CSI data, and extracting three data characteristics of average amplitude, phase difference and CIR amplitude distribution central moment from the CSI data; constructing the data features into images, and mapping the position information and the feature images by using a convolutional neural network; training a convolutional neural network using a dataset of images and class labels; and performing online positioning by using a convolutional network after training, and realizing position estimation by using a multi-image positioning algorithm based on spectral clustering. However, the CSI indoor positioning technology based on convolutional neural networks (Convolutional Neural Network, CNN) does not utilize topology information among access points, but simply splices CSI features of different access points into images, and then inputs a model for training and prediction. This approach results in loss of access point topology information, thereby limiting positioning accuracy.
Disclosure of Invention
The invention aims to provide a CSI indoor positioning method and device based on a graph neural network, which are used for performing indoor positioning based on CSI, and can fully utilize amplitude information and phase information of the CSI while maintaining topology information of an access point.
In order to achieve the above purpose, the technical scheme provided by the invention is as follows:
in a first aspect, the present invention provides a CSI indoor positioning method based on a graph neural network, including the following steps:
step 1: constructing an indoor positioning scene, comprising: arranging a plurality of access points for transmitting signals in an indoor positioning scene, setting a positioning area in the indoor positioning scene, and arranging a plurality of reference points for representing the position of a user terminal in the positioning area;
Step 2: the user terminal receives the transmitting signals from a plurality of access points at different reference points, extracts CSI fingerprint features respectively, performs dimension reduction processing on the CSI fingerprint features to obtain node features, wherein the node features comprise amplitude features and phase features, obtains edge features and adjacent matrixes according to topology information among the access points, constructs a graph structure corresponding to the CSI fingerprint features according to the node features, the edge features and the adjacent matrixes, marks graph class labels, and sets the graph structure and the graph class labels corresponding to the graph structure as a graph dataset;
Step 3: offline training of CSI-GNN algorithm models containing GNN1, GNN2, and MLP using graph dataset, comprising: inputting amplitude features into GNN1, inputting phase features into GNN2, inputting edge features and adjacent matrixes into GNN1 and GNN2, inputting MLP after splicing output features of the GNN1 and the GNN2, dynamically adjusting weight parameters by the MLP, outputting a prediction result, and obtaining a trained CSI-GNN algorithm model by minimizing a loss function between the prediction result and a graph class label or reaching the maximum iteration number;
step 4: and adopting a trained CSI-GNN algorithm model to perform online positioning and outputting the real-time position of the user terminal.
The technical conception of the invention is as follows: constructing an indoor positioning scene containing an access point and a user terminal position reference point; and when the user terminal is positioned at any one reference point, receiving the transmitting signals from all access points in the indoor positioning scene by the user terminal, extracting CSI fingerprint features from the transmitting signals, performing dimension reduction processing to obtain amplitude information and phase information of the transmitting signals respectively, and constructing feature vectors containing the two types of information, namely amplitude features and phase features, as node features of the corresponding access points.
Meanwhile, the invention further considers the significance of the topology information among the access points in the indoor scene on the positioning precision, so the invention obtains the edge weight and the adjacent matrix according to the topology information among the access points, the edge weight is used as the edge feature, the adjacent matrix is used for representing the adjacent relation of the nodes in the graph structure, the graph structure is constructed according to the node feature, the edge feature and the adjacent matrix, the corresponding graph class labels are obtained, the graph class labels are used for representing the reference points where the user terminal is located, and the graph structure and the corresponding graph class labels are collected into a graph data set.
Then, training a CSI-GNN algorithm model by adopting a graph data set, and training the CSI-GNN algorithm model by minimizing a cross entropy loss function; and using the trained CSI-GNN algorithm model for actual reasoning.
Further, in step 2, the CSI fingerprint is expressed as:
Where CSI AP represents a set of CSI fingerprints from N AP access points obtained at a certain reference point, CSI j is an nxmxk matrix, representing CSI fingerprints between a user terminal and a j-th access point when the user terminal is located at the current reference point, j=1, 2.
Further, in step 2, the dimension reduction processing is performed on the CSI fingerprint feature to obtain a node feature, where the node feature includes an amplitude feature and a phase feature, and specifically includes:
performing dimension reduction on the csi j to obtain a k-dimension vector, and expressing the k-dimension vector as follows:
wherein, L j represents a k-dimensional vector corresponding to a j-th node, L jt represents the amplitude of a t-th element of L j, and θ jt represents the phase of a t-th element of L j;
the amplitude characteristic and the phase characteristic are extracted from the k-dimensional vector and expressed as:
Wherein, LA represents amplitude characteristic, LP represents phase characteristic, LA j＝[lj1,lj2,...,ljk, which corresponds to the amplitude information of csi j, LP j＝[θj1,θj2,...,θjk, which corresponds to the phase information of csi j.
Further, in step2, the obtaining the edge feature according to the topology information between the access points specifically includes:
the topology information adopts the physical distance between access points;
obtaining edge weights according to the physical distance between the access points, and expressing the edge weights as follows:
Where w ij represents the edge weight, d ij represents the physical distance between the i-th access point and the j-th access point;
Edge weights are used as edge features.
Further, in step 2, based on the obtained edge weights, the topology information adopts the adjacency relation between the access points to construct an adjacency matrix, which is expressed as follows by a formula:
where a denotes an adjacency matrix, a ij denotes an edge feature from the ith node to the jth node, a ij＝wij when i+.j, and a ij＝aii =0 when i=j.
Further, in step 2, the construction of the graph structure corresponding to the CSI fingerprint feature according to the node feature, the edge feature and the adjacency matrix specifically includes:
Assuming that the user terminal is currently located at a certain reference point, receiving transmitting signals from N AP access points in an indoor positioning scene, wherein N AP CSI fingerprint features are corresponding to the transmitting signals;
N AP access points are used as nodes, amplitude features and phase features corresponding to the access points are used as node features, edge weights are used as edge features, the adjacent relation of the nodes is represented by an adjacent matrix A, and a graph structure corresponding to all CSI fingerprint features at the current reference point is obtained.
Further, in step 2, the graph class label specifically includes:
Y∈{0,1,...,NRP-1}
wherein Y represents a graph class label, each graph structure has only one graph class label, and assuming that the current user terminal is located at the kth reference point, the graph class label corresponding to the graph structure at the reference point is y=k-1.
Further, in step 3, after the output features of GNN1 and GNN2 are spliced, the output features are input into an MLP, the MLP dynamically adjusts weight parameters, and outputs a prediction result, which specifically includes:
The output characteristic of the GNN1 is a vector r 1, the output characteristic of the GNN2 is a vector r 2, the vector r 1 and the vector r 2 respectively correspond to the amplitude characteristic of the input GNN1 and the phase characteristic of the input GNN2, and the vector r 1 and the vector r 2 are spliced and then input into the MLP;
In the indoor positioning scene, based on the difference of the influence degree of noise on the amplitude characteristic and the phase characteristic, the MLP dynamically adjusts the weights of the vector r 1 corresponding to the amplitude characteristic and the vector r 2 corresponding to the phase characteristic in the prediction result, and the final prediction result is output by giving larger weight to the vector corresponding to the characteristic less influenced by the noise.
In a second aspect, in order to achieve the above object, an embodiment of the present invention further provides a CSI indoor positioning device based on a graph neural network, which includes an indoor positioning scene construction module, a graph dataset acquisition module, a CSI-GNN algorithm model training module, and a CSI-GNN algorithm model actual reasoning module;
The indoor positioning scene construction module is used for constructing an indoor positioning scene, and comprises: arranging a plurality of access points for transmitting signals in an indoor positioning scene, setting a positioning area in the indoor positioning scene, and arranging a plurality of reference points for representing the position of a user terminal in the positioning area;
The graph data set acquisition module is used for receiving transmitting signals from a plurality of access points at different reference points by a user terminal, respectively extracting CSI fingerprint features, performing dimension reduction processing on the CSI fingerprint features to obtain node features, wherein the node features comprise amplitude features and phase features, obtaining edge features and an adjacent matrix according to topology information among the access points, constructing a graph structure corresponding to the CSI fingerprint features and labeling graph class labels according to the node features, the edge features and the adjacent matrix, and collecting the graph structure and the graph class labels corresponding to the graph structure as a graph data set;
The CSI-GNN algorithm model training module is used for offline training a CSI-GNN algorithm model comprising GNN1, GNN2 and MLP by using a graph dataset, and comprises the following steps: inputting amplitude features into GNN1, inputting phase features into GNN2, inputting edge features and adjacent matrixes into GNN1 and GNN2, inputting MLP after splicing output features of the GNN1 and the GNN2, dynamically adjusting weight parameters by the MLP, outputting a prediction result, and obtaining a trained CSI-GNN algorithm model by minimizing a loss function between the prediction result and a graph class label or reaching the maximum iteration number;
the CSI-GNN algorithm model actual reasoning module is used for carrying out online positioning by adopting the trained CSI-GNN algorithm model and outputting the real-time position of the user terminal.
In order to achieve the above object, the embodiment of the present invention further provides a CSI indoor positioning device based on a graph neural network, which includes a memory and a processor, where the memory is configured to store a computer program, and the processor is configured to implement, when executing the computer program, the CSI indoor positioning method based on the graph neural network provided by the embodiment of the present invention in the first aspect.
The beneficial effects of the invention are as follows:
(1) The CSI-GNN algorithm model provided by the invention is used for indoor positioning based on the CSI, and compared with the RSSI, the CSI has a plurality of remarkable advantages. First CSI provides more detailed information about the wireless channel, including amplitude and phase variations of the signal. Such information can more accurately depict the propagation path of the signal in a complex indoor environment. For example, the phase information may be used to identify and distinguish multipath, which is not achievable by RSSI. Since RSSI provides only an overall measure of signal strength, the interaction of the signal and the environment cannot be carefully reflected. Second, the high dimensional nature of CSI makes it more efficient at processing signals in multipath propagation and non-line-of-sight conditions. Furthermore, the CSI provides more comprehensive information, which makes it more effective in machine learning and deep learning based indoor positioning systems that can utilize detailed information of CSI for more complex feature extraction and environmental understanding.
(2) The CSI-GNN algorithm model provided by the invention has excellent self-adaptive capacity, and particularly, when the amplitude characteristic or the phase characteristic is interfered by larger noise, the CSI-GNN algorithm model preferably uses the characteristic less influenced by the noise by dynamically adjusting the weight parameter of the MLP, thereby not only improving the accuracy and the stability of a positioning system, but also enhancing the robustness of the CSI-GNN algorithm model in the presence of changeable environments and complex noise.
(3) The CSI-GNN algorithm model processes the indoor positioning problem from the view point of the graph, and fully utilizes the topology information among the access points. According to the wireless communication theory, when the position of the user terminal is fixed, if the distance between two access points is closer, the CSI between the two access points and the user terminal is more similar. This similarity information is translated into edge weights for the graph. The graph convolution layer utilizes edge weights to aggregate node characteristics, so that adjacent nodes have more similar characteristics. Compared with the traditional CNN algorithm, the CSI-GNN fully utilizes the CSI information, and meanwhile, the loss of topology information between access points is avoided, so that the indoor positioning accuracy is further improved.
Drawings
Fig. 1 is a flowchart of a CSI indoor positioning method based on a neural network according to an embodiment of the present invention.
Fig. 2 is a specific flowchart of a CSI indoor positioning method based on a neural network according to an embodiment of the present invention.
Fig. 3 is a schematic diagram of an indoor positioning scene provided by an embodiment of the present invention.
Fig. 4 is a schematic structural diagram of a CSI-GNN algorithm according to an embodiment of the present invention.
Fig. 5 is a schematic structural diagram of GNN in CSI-GNN algorithm according to an embodiment of the present invention.
Fig. 6 is a schematic structural diagram of MLP in CSI-GNN algorithm according to an embodiment of the present invention.
Fig. 7 is a schematic structural diagram of a CSI indoor positioning device based on a neural network according to an embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the detailed description is presented by way of example only and is not intended to limit the scope of the invention.
As shown in fig. 1 and fig. 2, an embodiment provides a CSI indoor positioning method based on a graph neural network, which includes the following steps:
S110, constructing an indoor positioning scene, which comprises the following steps: a plurality of access points for transmitting signals are arranged in an indoor positioning scene, a positioning area is arranged in the indoor positioning scene, and a plurality of reference points for representing the position of a user terminal are arranged in the positioning area.
First, an indoor positioning scene is constructed, and in this embodiment, an indoor single-user positioning scene based on CSI is taken as an example to be described. As shown in fig. 3, N AP access points are distributed in the indoor single-user positioning scene, N RP reference points are set in the positioning area, and the physical positions of the reference points are known. A user is located at a reference point in the location area and carries a terminal device (e.g., a cell phone). The indoor positioning problem is described as: and judging which reference point the user is positioned according to the CSI fingerprint characteristics between the terminal equipment carried by the user and the N AP access points, and further determining the physical position of the user.
S120, the user terminal receives the transmitting signals from a plurality of access points at different reference points, extracts CSI fingerprint features respectively, performs dimension reduction processing on the CSI fingerprint features to obtain node features, wherein the node features comprise amplitude features and phase features, obtains edge features and adjacent matrixes according to topology information among the access points, constructs a graph structure corresponding to the CSI fingerprint features according to the node features, the edge features and the adjacent matrixes, marks graph class labels, and sets the graph structure and the graph class labels corresponding to the graph structure as a graph dataset.
When the ue is located at any reference point, CSI fingerprint features between the ue and N AP access points may be measured and obtained, expressed as:
Where CSI AP represents the set of CSI fingerprints from N AP access points obtained at a certain reference point, CSI j is a matrix of nxmxk, representing CSI fingerprints between the user terminal and the jth access point when the user is at the current reference point, j=1, 2.
In the indoor positioning scene, changing the position of the user terminal, and collecting CSI fingerprint features from N RP reference points. The CSI fingerprint at each reference point contains signal characteristics specific to that reference point, such as: amplitude information and phase information of the signal. And constructing the CSI fingerprint characteristics collected at each reference point into a fingerprint database.
For the CSI fingerprint features in the fingerprint database, amplitude features and phase features corresponding to the transmitting signals of the access points received at a certain reference point are extracted from the CSI fingerprint features to serve as node features, topology information among the access points is considered, physical distances among the access points are converted into edge weights, the edge weights serve as edge features, a graph structure corresponding to the CSI fingerprint features at each reference point is built according to the node features, the edge features and the adjacent matrix, in the embodiment, a full-connection undirected graph is specifically adopted for the graph structure, and a set of full-connection undirected graphs corresponding to the CSI fingerprint features collected at N RP reference points is a graph dataset. The construction of the full-connection undirected graph specifically comprises the following steps:
And taking the access points as nodes, wherein the number of the nodes is N AP, and when the user terminal is positioned at a certain reference point, the CSI fingerprint characteristic between the jth node and the user terminal is CSI j. Reducing the dimension of csi j into a k-dimensional vector Two new vectors la j＝[lj1,lj2,...,ljk and lp j＝[θj1,θj2,...,θjk are thus obtained, where la j corresponds to the amplitude information of csi j and lp j corresponds to the phase information of csi j. la j and lp j respectively constitute two classes of node features for the j-th node, namely: amplitude characteristicsAnd phase characteristics
The location of the access point in the indoor positioning scenario is fixed. The physical distance between access points is converted into edge weights of the full connection undirected graph as edge features. The edge weight w ij between the i-th node and the j-th node is represented as follows:
Where d ij denotes the physical distance between the i-th access point and the j-th access point.
The adjacency matrix A of the full-connection undirected graph reflects the adjacency relation among the nodes, and is constructed according to the adjacency relation among the nodes and the edge weight and expressed as follows by a formula:
Wherein, when i+.j, a ij＝wij. When i+.j, a ij＝aii =0.
After the graph structure is expressed, the indoor positioning problem is converted into a graph classification problem, namely: the category of the graph is judged according to the graph structure information (node characteristics, edge characteristics and adjacency matrix). The number of categories is the same as the number of reference points, N RP. Define a graph class label Y e {0,1,.., N RP -1}, there is one and only one graph class label per fully connected undirected graph. Assuming that the user is located at the kth reference point during sampling, the graph class label corresponding to the full-connection undirected graph at the reference point is y=k-1.
S130, training a CSI-GNN algorithm model comprising GNN1, GNN2 and MLP offline by using a graph dataset, wherein the method comprises the following steps: inputting the amplitude characteristic into the GNN1, inputting the phase characteristic into the GNN2, inputting the edge characteristic and the adjacent matrix into the GNN1 and the GNN2, inputting the MLP after the output characteristics of the GNN1 and the GNN2 are spliced, dynamically adjusting the weight parameter by the MLP, outputting the prediction result, and obtaining the trained CSI-GNN algorithm model by minimizing the loss function between the prediction result and the graph class label or reaching the maximum iteration number.
The structural block diagram of the CSI-GNN algorithm model designed by the invention is shown in figure 4. The indoor positioning method based on the CSI-GNN is divided into an offline stage and an online stage. In S120, the construction of the graph dataset is completed, and then the graph dataset is used to perform offline training on the CSI-GNN algorithm model, and the related steps are described as follows:
1): and respectively taking the amplitude characteristic LA and the phase characteristic LP as node characteristics of the GNN1 and the GNN2, taking the edge weight of the adjacent matrix A as the edge characteristics of the GNN1 and the GNN2, and carrying out forward propagation on the CSI-GNN algorithm model.
GNN1 and GNN2 are GNNs identical in structure, and the structure is shown in fig. 5. Each GNN contains three GraphConv graph convolutional layers, the inputs to each graph convolutional layer including: node characteristics, edge index, edge weight. The edge index indicates the connection relationship between the nodes. The edge index and edge weight are derived from the adjacency matrix a. The graph convolution layer updates node characteristics by aggregating the characteristics of nodes and their neighbors. The manner of aggregation of the GraphConv layers of the convolution is as follows:
where x i represents the node characteristics of node i, { x j |j ε N (i) } represents all neighbor nodes of node i. a ji is an element in the adjacency matrix a. W 1 and W 2 are learnable parameters of the model. x i' is the node characteristics of node i after being updated by the GraphConv graph convolutional layer.
Each GraphConv is followed by one BatchNorm layer. For adjusting and normalizing the layer output to improve the stability and learning efficiency of the model. Using a ReLU activation function after each BatchNorm layers introduces nonlinearities that enhance the model's ability to process complex data. Finally, the features of all nodes are processed by global average pooling and are summarized into a graph-level feature representation for graph classification tasks. The principle of global average pooling is as follows:
the global averaging pooling is essentially that all node features of each fully connected undirected graph are averaged, so that a feature vector r representing the graph-level feature is obtained, and the feature vector r is the output feature of the GNN.
2): The output characteristics of the GNN1 and the GNN2 are spliced and then input into an MLP, wherein the MLP comprises three linear layers, and a ReLU activation function is added between the adjacent linear layers. The structure of the MLP is shown in fig. 6. The output characteristics of the MLP are denoted Logits. Logits is an N RP -dimensional vector, where the index of the maximum value in the vector is p, which indicates that the CSI-GNN algorithm model predicts that the user terminal is located at the reference point corresponding to the graph class label p.
3): And calculating Logits cross entropy loss between the graph class label Y, carrying out back propagation, calculating gradient according to the cross entropy loss, and updating the learnable parameters of the model through an Adam optimizer.
4): And repeatedly executing forward propagation and backward propagation until the cross entropy loss is smaller than a set threshold value or reaches a set iteration number, and obtaining a trained CSI-GNN algorithm model for real-time indoor positioning prediction.
And S140, performing online positioning by adopting the trained CSI-GNN algorithm model, and outputting the real-time position of the user terminal.
The trained CSI-GNN algorithm model is used for positioning tasks of indoor scenes and any complex scenes, and the CSI-GNN algorithm model provided by the invention not only considers amplitude information and phase information of CSI, but also considers topology information of an access point. When the amplitude characteristic or the phase characteristic of the CSI is interfered by larger noise, the method provided by the invention can be used for preferentially using the characteristic less affected by the noise by dynamically adjusting the weight parameter of the MLP, thereby improving the accuracy and the robustness of the positioning system. Therefore, the invention can obtain better effect not only in indoor positioning scenes, but also in similar complex scenes, thereby ensuring good detection effect.
Based on the same inventive concept, the embodiment of the invention also provides a CSI indoor positioning device 700 based on a graph neural network, as shown in fig. 7, comprising an indoor positioning scene construction module 710, a graph dataset acquisition module 720, a CSI-GNN algorithm model training module 730 and a CSI-GNN algorithm model actual reasoning module 740;
The indoor positioning scene construction module 710 is configured to construct an indoor positioning scene, including: arranging a plurality of access points for transmitting signals in an indoor positioning scene, setting a positioning area in the indoor positioning scene, and arranging a plurality of reference points for representing the position of a user terminal in the positioning area;
The graph dataset acquisition module 720 is configured to receive transmitting signals from a plurality of access points at different reference points and extract CSI fingerprint features respectively, perform dimension reduction processing on the CSI fingerprint features to obtain node features, where the node features include amplitude features and phase features, obtain edge features and an adjacency matrix according to topology information between the access points, and construct a graph structure corresponding to the CSI fingerprint features and label graph class labels according to the node features, the edge features and the adjacency matrix, where a set of the graph structure and the graph class labels corresponding to the graph structure is a graph dataset;
The CSI-GNN algorithm model training module 730 is configured to train the CSI-GNN algorithm model including GNN1, GNN2, and MLP offline using the graph dataset, and includes: inputting amplitude features into GNN1, inputting phase features into GNN2, inputting edge features and adjacent matrixes into GNN1 and GNN2, inputting MLP after splicing output features of the GNN1 and the GNN2, dynamically adjusting weight parameters by the MLP, outputting a prediction result, and obtaining a trained CSI-GNN algorithm model by minimizing a loss function between the prediction result and a graph class label or reaching the maximum iteration number;
The CSI-GNN algorithm model actual inference module 740 is configured to perform online positioning by using the trained CSI-GNN algorithm model, and output a real-time position of the user terminal.
Based on the same inventive concept, the embodiment also provides a CSI indoor positioning device based on the graph neural network, which comprises a memory and a processor, wherein the memory is used for storing a computer program, and the processor is used for realizing the CSI indoor positioning method based on the graph neural network when executing the computer program.
It should be noted that, the CSI indoor positioning device based on the graph neural network and the CSI indoor positioning device based on the graph neural network provided in the foregoing embodiments are all the same as the CSI indoor positioning method embodiment based on the graph neural network, and specific implementation processes of the CSI indoor positioning device based on the graph neural network are detailed in the CSI indoor positioning method embodiment based on the graph neural network, which is not repeated herein.
The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the foregoing detailed description of the invention has been provided, it will be apparent to those skilled in the art that modifications may be made to the embodiments described in the foregoing examples, and that certain features may be substituted for those illustrated and described herein. Modifications, equivalents, and alternatives falling within the spirit and principles of the invention are intended to be included within the scope of the invention.

Claims (10)

1.一种基于图神经网络的CSI室内定位方法，其特征在于，包括以下步骤：1. A CSI indoor positioning method based on graph neural networks, characterized by comprising the following steps: 步骤1：构建室内定位场景，包括：在室内定位场景中布置若干个用于发射信号的接入点，在室内定位场景中设置定位区域，在定位区域中布置若干个用于表征用户终端位置的参考点；Step 1: Construct an indoor positioning scenario, including: setting up several access points for transmitting signals in the indoor positioning scenario, setting up a positioning area in the indoor positioning scenario, and setting up several reference points in the positioning area to represent the location of the user terminal. 步骤2：用户终端在不同参考点处接收来自若干个接入点的发射信号并分别提取CSI指纹特征，对CSI指纹特征进行降维处理得到节点特征，所述节点特征包含幅度特征和相位特征，根据接入点之间的拓扑信息得到边特征和邻接矩阵，根据节点特征、边特征以及邻接矩阵构建CSI指纹特征对应的图结构并标注图类别标签，图结构及其对应的图类别标签的集合为图数据集；Step 2: The user terminal receives transmitted signals from several access points at different reference points and extracts CSI fingerprint features respectively. The CSI fingerprint features are dimensionality reduced to obtain node features, which include amplitude features and phase features. Edge features and adjacency matrix are obtained based on the topological information between access points. The graph structure corresponding to the CSI fingerprint features is constructed based on the node features, edge features and adjacency matrix and the graph category label is labeled. The set of graph structure and its corresponding graph category label is the graph dataset. 步骤3：利用图数据集离线训练包含GNN1、GNN2以及MLP的CSI-GNN算法模型，包括：将幅度特征输入GNN1，将相位特征输入GNN2，将边特征和邻接矩阵输入GNN1和GNN2，GNN1和GNN2的输出特征拼接后输入MLP，MLP动态调整权重参数，输出预测结果，通过预测结果与图类别标签之间的损失函数最小化，或达到最大迭代次数，得到训练好的CSI-GNN算法模型；Step 3: Train the CSI-GNN algorithm model offline using the graph dataset, which includes GNN1, GNN2 and MLP. This includes: inputting amplitude features into GNN1, phase features into GNN2, edge features and adjacency matrix into GNN1 and GNN2, concatenating the output features of GNN1 and GNN2 and inputting them into MLP, dynamically adjusting the weight parameters of MLP, outputting the prediction results, and minimizing the loss function between the prediction results and the graph category labels, or reaching the maximum number of iterations, to obtain the trained CSI-GNN algorithm model. 步骤4：采用训练好的CSI-GNN算法模型，进行在线定位，输出用户终端的实时位置。Step 4: Use the trained CSI-GNN algorithm model to perform online localization and output the real-time location of the user terminal. 2.根据权利要求1所述的基于图神经网络的CSI室内定位方法，其特征在于，步骤2中，所述的CSI指纹特征，用公式表示为：2. The CSI indoor positioning method based on graph neural networks according to claim 1, characterized in that, in step 2, the CSI fingerprint feature is expressed by the formula: 其中，CSIAP表示某一参考点处获得的来自NAP个接入点的CSI指纹特征的集合，csij是一个N×M×K的矩阵，表示用户终端位于当前参考点时与第j个接入点之间的CSI指纹特征，j＝1,2,...,NAP，N表示用户终端的接收天线数量，M表示每个接入点的发射天线数量，K表示子载波数。Wherein, CSI AP represents the set of CSI fingerprint features obtained from N AP access points at a certain reference point, and csi j is an N×M×K matrix representing the CSI fingerprint features between the user terminal at the current reference point and the j-th access point, j＝1,2,...,N AP , where N represents the number of receiving antennas of the user terminal, M represents the number of transmitting antennas of each access point, and K represents the number of subcarriers. 3.根据权利要求2所述的基于图神经网络的CSI室内定位方法，其特征在于，步骤2中，所述的对CSI指纹特征进行降维处理得到节点特征，所述节点特征包含幅度特征和相位特征，具体为：3. The CSI indoor positioning method based on graph neural networks according to claim 2, characterized in that, in step 2, the dimensionality reduction processing of the CSI fingerprint features to obtain node features, wherein the node features include amplitude features and phase features, specifically: 对csij进行降维处理，得到k维向量，用公式表示为：Dimensionality reduction of csi j yields a k-dimensional vector, expressed by the formula: 其中，Lj表示第j个节点对应的k维向量，ljt表示Lj第t个元素的幅度，θjt表示Lj第t个元素的相位；Where L <sub>j</sub> represents the k-dimensional vector corresponding to the j-th node, l <sub>jt</sub> represents the magnitude of the t-th element of L<sub> j </sub>, and θ <sub>jt</sub> represents the phase of the t-th element of L<sub> j </sub>. 从k维向量中提取得到幅度特征和相位特征，用公式表示为：The amplitude and phase features are extracted from the k-dimensional vector, expressed by the following formula: 其中，LA表示幅度特征，LP表示相位特征，laj＝[lj1,lj2,...,ljk]，对应csij的幅度信息，lpj＝[θj1,θj2,...,θjk]，对应csij的相位信息。Where LA represents amplitude characteristics, LP represents phase characteristics, la j = [l j1 ,l j2 ,...,l jk ], corresponding to the amplitude information of csi j , and lp j = [θ j1 ,θ j2 ,...,θ jk ], corresponding to the phase information of csi j . 4.根据权利要求1所述的基于图神经网络的CSI室内定位方法，其特征在于，步骤2中，所述的根据接入点之间的拓扑信息得到边特征，具体为：4. The CSI indoor positioning method based on graph neural networks according to claim 1, characterized in that, in step 2, obtaining edge features based on the topological information between access points specifically includes: 所述拓扑信息采用接入点之间的物理距离；The topology information is based on the physical distance between access points; 根据接入点之间的物理距离得到边权重，用公式表示为：The edge weights are obtained based on the physical distance between access points, and can be expressed by the following formula: 其中，wij表示边权重，dij表示第i个接入点和第j个接入点之间的物理距离；Where w<sub>ij</sub> represents the edge weight, and d <sub>ij</sub> represents the physical distance between the i-th access point and the j-th access point; 采用边权重作为边特征。Edge weights are used as edge features. 5.根据权利要求4所述的基于图神经网络的CSI室内定位方法，其特征在于，步骤2中，基于得到的边权重，拓扑信息采用接入点之间的邻接关系，构建邻接矩阵，用公式表示为：5. The CSI indoor positioning method based on graph neural networks according to claim 4, characterized in that, in step 2, based on the obtained edge weights, the topology information uses the adjacency relationship between access points to construct an adjacency matrix, expressed by the formula: 其中，A表示邻接矩阵，aij表示从第i个节点到第j个节点的边特征，当i≠j时，aij＝wij，当i＝j时，aij＝aii＝0。Where A represents the adjacency matrix, a <sub>ij</sub> represents the edge feature from the i-th node to the j-th node, when i ≠ j, a <sub>ij</sub> = w <sub>ij</sub> , when i = j, a<sub>ij</sub> = a <sub>ii </sub> = 0. 6.根据权利要求3或4或5所述的基于图神经网络的CSI室内定位方法，其特征在于，步骤2中，所述的根据节点特征、边特征以及邻接矩阵构建CSI指纹特征对应的图结构，具体为：6. The CSI indoor positioning method based on graph neural networks according to claim 3, 4, or 5, characterized in that, in step 2, the construction of the graph structure corresponding to the CSI fingerprint features based on node features, edge features, and adjacency matrix specifically includes: 假设用户终端当前位于某一参考点处，接收到来自室内定位场景中NAP个接入点的发射信号，对应有NAP个CSI指纹特征；Assume that the user terminal is currently located at a certain reference point and receives transmission signals from N AP access points in the indoor positioning scenario, which correspond to N AP CSI fingerprint features; 以NAP个接入点为节点，以接入点对应的幅度特征和相位特征为节点特征，以边权重为边特征，节点的邻接关系由邻接矩阵A表征，得到当前参考点处的所有CSI指纹特征对应的图结构。Using N AP access points as nodes, the amplitude and phase features corresponding to the access points as node features, and the edge weights as edge features, the adjacency relationship of the nodes is represented by the adjacency matrix A, thus obtaining the graph structure corresponding to all CSI fingerprint features at the current reference point. 7.根据权利要求1所述的基于图神经网络的CSI室内定位方法，其特征在于，步骤2中，所述的图类别标签，具体为：7. The CSI indoor positioning method based on graph neural networks according to claim 1, characterized in that, in step 2, the graph category label specifically comprises: Y∈{0,1,...,NRP-1}Y∈{0,1,...,N RP -1} 其中，Y表示图类别标签，每个图结构有且仅有一个图类别标签，假设当前用户终端位于第k个参考点，则该参考点处图结构对应的图类别标签为Y＝k-1。Where Y represents the graph category label, and each graph structure has one and only one graph category label. Assuming that the current user terminal is located at the k-th reference point, the graph category label corresponding to the graph structure at that reference point is Y = k-1. 8.根据权利要求7所述的基于图神经网络的CSI室内定位方法，其特征在于，步骤3中，所述的将GNN1和GNN2的输出特征进行拼接后输入MLP，MLP动态调整权重参数，输出预测结果，具体为：8. The CSI indoor positioning method based on graph neural networks according to claim 7, characterized in that, in step 3, the concatenation of the output features of GNN1 and GNN2 and input into the MLP, the dynamic adjustment of the weight parameters by the MLP, and the output of the prediction result are specifically as follows: GNN1的输出特征为向量r1，GNN2的输出特征为向量r2，向量r1和向量r2分别对应输入GNN1的幅度特征和输入GNN2的相位特征，将向量r1和向量r2拼接后输入MLP；The output feature of GNN1 is vector r1 , and the output feature of GNN2 is vector r2. Vector r1 and vector r2 correspond to the magnitude feature of input GNN1 and the phase feature of input GNN2, respectively. Vector r1 and vector r2 are concatenated and then input into MLP. 室内定位场景中，基于幅度特征和相位特征受噪声影响程度的不同，MLP动态调整幅度特征对应的向量r1和相位特征对应的向量r2在预测结果中的权重，通过赋予受噪声影响较小的特征对应的向量以较大的权重，输出最终的预测结果。In indoor positioning scenarios, based on the different degrees to which amplitude and phase features are affected by noise, the MLP dynamically adjusts the weights of the vector r1 corresponding to the amplitude feature and the vector r2 corresponding to the phase feature in the prediction result. By assigning a larger weight to the vector corresponding to the feature less affected by noise, the final prediction result is output. 9.一种基于图神经网络的CSI室内定位装置，其特征在于，包括室内定位场景构建模块、图数据集获取模块、CSI-GNN算法模型训练模块、CSI-GNN算法模型实际推理模块；9. A CSI indoor positioning device based on graph neural networks, characterized in that it includes an indoor positioning scene construction module, a graph dataset acquisition module, a CSI-GNN algorithm model training module, and a CSI-GNN algorithm model actual inference module; 所述室内定位场景构建模块用于构建室内定位场景，包括：在室内定位场景中布置若干个用于发射信号的接入点，在室内定位场景中设置定位区域，在定位区域中布置若干个用于表征用户终端位置的参考点；The indoor positioning scene construction module is used to construct an indoor positioning scene, including: arranging a number of access points for transmitting signals in the indoor positioning scene, setting a positioning area in the indoor positioning scene, and arranging a number of reference points for representing the location of the user terminal in the positioning area. 所述图数据集获取模块用于用户终端在不同参考点处接收来自若干个接入点的发射信号并分别提取CSI指纹特征，对CSI指纹特征进行降维处理得到节点特征，所述节点特征包含幅度特征和相位特征，根据接入点之间的拓扑信息得到边特征和邻接矩阵，根据节点特征、边特征以及邻接矩阵构建CSI指纹特征对应的图结构并标注图类别标签，图结构及其对应的图类别标签的集合为图数据集；The graph dataset acquisition module is used by the user terminal to receive transmitted signals from several access points at different reference points and extract CSI fingerprint features respectively. The CSI fingerprint features are then dimensionality-reduced to obtain node features, which include amplitude features and phase features. Edge features and adjacency matrices are obtained based on the topological information between access points. The graph structure corresponding to the CSI fingerprint features is constructed based on the node features, edge features, and adjacency matrix, and graph category labels are labeled. The set of graph structures and their corresponding graph category labels is the graph dataset. 所述CSI-GNN算法模型训练模块用于利用图数据集离线训练包含GNN1、GNN2以及MLP的CSI-GNN算法模型，包括：将幅度特征输入GNN1，将相位特征输入GNN2，将边特征和邻接矩阵输入GNN1和GNN2，GNN1和GNN2的输出特征拼接后输入MLP，MLP动态调整权重参数，输出预测结果，通过预测结果与图类别标签之间的损失函数最小化，或达到最大迭代次数，得到训练好的CSI-GNN算法模型；The CSI-GNN algorithm model training module is used to train the CSI-GNN algorithm model containing GNN1, GNN2 and MLP offline using a graph dataset. The process includes: inputting amplitude features into GNN1, inputting phase features into GNN2, inputting edge features and adjacency matrix into GNN1 and GNN2, concatenating the output features of GNN1 and GNN2 and inputting them into MLP, dynamically adjusting the weight parameters of MLP, outputting the prediction result, and minimizing the loss function between the prediction result and the graph category label, or reaching the maximum number of iterations, to obtain the trained CSI-GNN algorithm model. 所述CSI-GNN算法模型实际推理模块用于采用训练好的CSI-GNN算法模型，进行在线定位，输出用户终端的实时位置。The CSI-GNN algorithm model's actual inference module is used to perform online positioning using the trained CSI-GNN algorithm model and output the real-time location of the user terminal. 10.一种基于图神经网络的CSI室内定位设备，包括存储器和处理器，所述存储器用于存储计算机程序，其特征在于，所述处理器用于当执行所述计算机程序时，实现权利要求1-8任一项所述的基于图神经网络的CSI室内定位方法。10. A CSI indoor positioning device based on a graph neural network, comprising a memory and a processor, wherein the memory is used to store a computer program, and the processor is used to implement the CSI indoor positioning method based on a graph neural network according to any one of claims 1-8 when executing the computer program.
CN202311830099.3A 2023-12-28 2023-12-28 CSI indoor positioning method and device based on graph neural network Pending CN117939402A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202311830099.3A CN117939402A (en) 2023-12-28 2023-12-28 CSI indoor positioning method and device based on graph neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202311830099.3A CN117939402A (en) 2023-12-28 2023-12-28 CSI indoor positioning method and device based on graph neural network

Publications (1)

Publication Number Publication Date
CN117939402A true CN117939402A (en) 2024-04-26

Family

ID=90749874

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202311830099.3A Pending CN117939402A (en) 2023-12-28 2023-12-28 CSI indoor positioning method and device based on graph neural network

Country Status (1)

Country Link
CN (1) CN117939402A (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN119255199A (en) * 2024-09-30 2025-01-03 北京中科创途科技有限公司 User location prediction method, device, terminal device and storage medium
CN119342414A (en) * 2024-10-21 2025-01-21 南京邮电大学 A channel state information indoor positioning method based on graph homogeneous network
CN119513612A (en) * 2025-01-17 2025-02-25 合肥综合性国家科学中心人工智能研究院(安徽省人工智能实验室) Training method of graph neural localization model, method and device for localizing target object
CN120351943A (en) * 2025-06-24 2025-07-22 长春职业技术学院 Vehicle navigation positioning method and system when navigation signal is lost

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN119255199A (en) * 2024-09-30 2025-01-03 北京中科创途科技有限公司 User location prediction method, device, terminal device and storage medium
CN119255199B (en) * 2024-09-30 2025-08-05 中网数安(北京)科技有限公司 User location prediction method, device, terminal device and storage medium
CN119342414A (en) * 2024-10-21 2025-01-21 南京邮电大学 A channel state information indoor positioning method based on graph homogeneous network
CN119342414B (en) * 2024-10-21 2025-09-26 南京邮电大学 A channel state information indoor positioning method based on graph homogeneous network
CN119513612A (en) * 2025-01-17 2025-02-25 合肥综合性国家科学中心人工智能研究院(安徽省人工智能实验室) Training method of graph neural localization model, method and device for localizing target object
CN119513612B (en) * 2025-01-17 2025-05-06 合肥综合性国家科学中心人工智能研究院(安徽省人工智能实验室) Training method of graph neural localization model, method and device for localizing target object
CN120351943A (en) * 2025-06-24 2025-07-22 长春职业技术学院 Vehicle navigation positioning method and system when navigation signal is lost

Similar Documents

Publication Publication Date Title
Zhu et al. BLS-location: A wireless fingerprint localization algorithm based on broad learning
CN117939402A (en) CSI indoor positioning method and device based on graph neural network
US10884112B2 (en) Fingerprint positioning method and system in smart classroom
CN111479231B (en) An indoor fingerprint localization method for millimeter-wave massive MIMO system
CN105790813B (en) Code book selection method based on deep learning under a kind of extensive MIMO
Wang et al. Graph-Neural-Network-Based WiFi indoor localization system with access point selection
CN107529222A (en) A kind of WiFi indoor locating systems based on deep learning
CN107576330A (en) A kind of localization method of the indoor dynamic sensing strategy based on WLAN fingerprints
CN113207089A (en) Position fingerprint positioning method based on CSI and crowdsourcing migration self-calibration updating
Chen et al. Smartphone-based indoor fingerprinting localization using channel state information
CN110351660B (en) A Bluetooth indoor positioning method based on two-step fingerprint matching architecture
CN115567871A (en) WiFi fingerprint indoor floor identification and position estimation method
Sun et al. A novel GCN based indoor localization system with multiple access points
CN110933628A (en) Fingerprint indoor positioning method based on twin network
Wang et al. Review of few-shot learning application in CSI human sensing
Chen et al. DeepMetricFi: Improving Wi-Fi fingerprinting localization by deep metric learning
Hsu et al. An adaptive Wi-Fi indoor localisation scheme using deep learning
Huang et al. MAPS: Indoor localization algorithm based on multiple AP selection
CN112887909B (en) Indoor positioning method based on Wi-Fi signals
CN109889975B (en) Terminal fingerprint positioning method based on NB-IoT
Rao et al. Device-free passive wireless localization system with transfer deep learning method
Li et al. Hierarchical Stagewise Training of Linked Deep Neural Networks for Multibuilding and Multifloor Indoor Localization Based on Wi-Fi RSSI Fingerprinting
CN115099385B (en) Spectrum map construction method based on sensor layout optimization and adaptive Kriging model
Gu et al. Graph convolutional network empowered indoor localization method via aggregating MIMO CSI
CN112488151B (en) A Semi-supervised Manifold Learning-Based Location Fingerprint Library Construction Method

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination