Dynamic scene indoor positioning method, device and equipment based on CSI data

Abstract

This application provides a dynamic scene indoor positioning method, apparatus, and device based on CSI data, which can be applied to the field of wireless positioning technology. The method includes: in response to the target device's transmission signal being blocked by at least one person in an indoor environment, acquiring raw measurement data of the access point; reducing the dimensionality of the complex matrix of the raw channel state information to obtain a flattened vector, thus obtaining preprocessed raw channel state information; inputting the preprocessed channel state information into a temporal convolutional network to generate hidden temporal features; processing the hidden temporal features using a first network branch and a second network branch to obtain dynamic and static features, respectively; concatenating the static features with the coordinate features of their corresponding access points to obtain location-aware static features; performing feature processing based on a Transformer module to obtain location-aware fusion features; and inputting the location-aware fusion features into a multilayer perceptron to obtain the predicted location.

Description

Technical Field

[0001] This application relates to the field of wireless positioning technology, and more specifically to a dynamic scene indoor positioning method, apparatus and equipment based on CSI data. Background Technology

[0002] In recent years, with the rapid development of wireless communication technology and the widespread adoption of mobile smart terminals, location-based services have become an important component of the modern information industry, widely applied in various fields of social development and possessing enormous application potential. Satellite positioning technology, represented by the Global Positioning System (GPS), has achieved all-weather coverage of outdoor environments worldwide, significantly changing people's production and lifestyles. However, satellite signals experience severe attenuation when penetrating buildings, making effective positioning difficult in indoor environments. As society's demand for location services continues to increase, the need for high-precision, real-time positioning in indoor environments is becoming increasingly urgent, driving the rapid development of indoor positioning technology.

[0003] Channel State Information (CSI) characterizes the multipath features of wireless signals under specific propagation conditions and is highly sensitive to environmental changes, making high-precision indoor positioning possible. However, in real-world applications, dynamic factors such as human activity, object movement, and the opening and closing of doors and windows continuously alter the propagation path of wireless signals, leading to significant changes in signal characteristics. This dynamic interference causes a mismatch between the CSI data collected during the training phase and the input features during the real-time positioning phase, resulting in decreased adaptability of the positioning model and significant positioning errors.

[0004] Currently, most existing indoor positioning research focuses on eliminating hardware biases by designing complex input features or optimizing network structure design to improve theoretical accuracy in specific environments. While these improvements have enhanced performance to some extent, they have not fundamentally solved the problem of insufficient model generalization ability caused by environmental dynamism. Therefore, maintaining high accuracy of CSI positioning in indoor environments with dynamic factors has become a key technical problem that urgently needs to be solved in this field. Summary of the Invention

[0005] The embodiments of this application provide a dynamic scene indoor positioning method, apparatus and equipment based on CSI data, which can at least solve the problem of low positioning accuracy caused by dynamic factors in the prior art.

[0006] According to the first aspect of this application, a dynamic scene indoor positioning method based on CSI data is provided, comprising: in response to the transmission signal of a target device in an indoor environment being blocked by at least one person, acquiring raw measurement data from N access points in the indoor environment, the raw measurement data including raw channel state information, wherein N is a positive integer greater than or equal to 1; performing dimensionality reduction processing on at least one complex matrix of the raw channel state information to obtain a flattened vector, thereby obtaining preprocessed raw channel state information; inputting the preprocessed channel state information into a temporal convolutional network to encode local temporal dependencies, generating hidden temporal features; and processing the hidden temporal features using a first network branch to obtain dynamic features, wherein the dynamic feature table is... The transmitted signal is reflected to N access points after being blocked by at least one person. A second network branch is used to process the hidden temporal features to obtain static features. These static features represent the transmitted signal reflected to N access points after being blocked by an object other than at least one person. The first and second network branches are trained based on different objective functions. The static features are concatenated with the coordinate features of their corresponding access points to obtain location-aware static features. The coordinate features characterize the position of the corresponding access point in the indoor environment. The location-aware static features are then processed using a Transformer module to obtain location-aware fusion features. These location-aware fusion features are then input into a multilayer perceptron to obtain the predicted position.

[0007] According to an embodiment of this application, concatenating static features with the coordinate features of their corresponding access points to obtain location-aware static features includes: when N is greater than or equal to 2, concatenating multiple static features with multiple coordinate features, wherein each static feature corresponds to an access point, and the multiple coordinate features represent multiple access points distributed in multiple different locations in the indoor environment.

[0008] According to an embodiment of this application, the method further includes: inputting the features obtained from the execution processes of the first network branch and the second network branch into a pre-constructed first loss function set, wherein the first loss function set includes a similarity loss function, an orthogonality loss function, and a reconstruction loss function; inputting the output of the multilayer perceptron into a pre-constructed second loss function set, wherein the second loss function set includes a prediction location loss function; performing weighted fusion of all loss function values ​​in the first loss function set and the second loss function set to obtain a total loss function value; and updating the parameters of the temporal convolutional network, the first network branch, the second network branch, the Transformer module, and the multilayer perceptron based on the total loss function value through gradient backpropagation.

[0009] According to embodiments of this application, a similarity loss function is used to minimize the differences between static features at consecutive time points; an orthogonal loss function is used to minimize the correlation between static features and dynamic features; a reconstruction loss function is used to minimize the difference between the sum of the reconstructed static and dynamic signals and the input signal; and a prediction position loss function is used to minimize the error between the predicted position and the true position.

[0010] According to an embodiment of this application, the method further includes: performing amplitude calibration on the preprocessed original channel state information to obtain a calibrated data matrix, wherein the data structure of the preprocessed original channel state information is in matrix form; and taking the square of the absolute value of each element in the calibrated data matrix to obtain the channel frequency response power matrix.

[0011] According to an embodiment of this application, amplitude calibration is performed on the preprocessed original channel state information to obtain a calibrated data matrix, which includes: calculating a scaling factor introduced by automatic gain control based on the received signal strength and the preprocessed original channel state information; and dividing each element of the preprocessed original channel state information by the scaling factor to obtain the calibrated data matrix.

[0012] According to an embodiment of this application, after concatenating the static features with the coordinate features of their corresponding access points to obtain location-aware static features, the method includes: adding a classification token as a prefix to the location-aware static features, wherein the classification token is used to identify feature category information at the front end of the location-aware static feature sequence.

[0013] According to an embodiment of this application, the method of obtaining location-aware fusion features by performing feature processing on location-aware static features based on Transformer modules includes: inputting location-aware static features into multiple Transformer modules cascaded in sequence for feature transformation processing, wherein each Transformer module performs spatial location information interaction and fusion on the location-aware static features through a self-attention mechanism; and generating location fusion features based on the features output by the multiple Transformer modules.

[0014] The second aspect of this application provides a dynamic scene indoor positioning device based on CSI data, comprising: a raw measurement data acquisition module, configured to acquire raw measurement data from N access points in an indoor environment in response to the transmission signal of a target device in an indoor environment being blocked by at least one person; the raw measurement data including raw channel state information, wherein N is a positive integer greater than or equal to 1; a channel state information preprocessing module, configured to perform dimensionality reduction processing on at least one complex matrix of the raw channel state information to obtain a flattened vector, thereby obtaining preprocessed raw channel state information; a time feature generation module, configured to input the preprocessed channel state information into a temporal convolutional network to encode local temporal dependencies, thereby generating hidden time features; and a dynamic feature extraction module, configured to process the hidden time features using a first network branch to obtain dynamic features, wherein the dynamic features represent the transmission signal. The system includes a signal that is reflected to N access points after being blocked by at least one person; a static feature extraction module, which processes hidden temporal features using a second network branch to obtain static features, which represent the signal reflected to N access points after being blocked by an object other than at least one person; a location-aware static feature generation module, which concatenates the static features with the coordinate features of their corresponding access points to obtain location-aware static features, where the coordinate features characterize the position of the corresponding access point in the indoor environment; a location-aware fusion feature acquisition module, which performs feature processing on the location-aware static features based on the Transformer module to obtain location-aware fusion features; and a predicted position determination module, which inputs the location-aware fusion features into a multilayer perceptron to obtain the predicted position.

[0015] A third aspect of this application provides an electronic device comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the method described above.

[0016] A fourth aspect of this application also provides a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implement the steps of the above-described method.

[0017] The fifth aspect of this application also provides a computer program product, including a computer program or instructions that, when executed by a processor, implement the steps of the above-described method.

[0018] According to the embodiments of this application, after extracting static and dynamic features from the channel state information, positioning is performed using only the extracted static features, which avoids the influence of dynamic features on positioning, thereby avoiding the resulting errors and improving the accuracy of positioning based on CSI data. Attached Figure Description

[0019] The above-mentioned contents, as well as other objects, features, and advantages of this application, will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0020] Figure 1 The illustration shows an application scenario of the dynamic scene indoor positioning method, apparatus and device based on CSI data according to embodiments of this application;

[0021] Figure 2 A flowchart illustrating a dynamic scene indoor positioning method based on CSI data according to an embodiment of this application is shown schematically.

[0022] Figure 3 A flowchart illustrating the setting of a loss function according to an embodiment of this application is shown schematically;

[0023] Figure 4 This illustration schematically shows another flowchart of a dynamic scene indoor positioning method based on CSI data according to an embodiment of this application;

[0024] Figure 5A A tabular diagram illustrating experimental scenarios according to embodiments of this application is shown schematically.

[0025] Figure 5B A tabular diagram illustrating experimental methods according to embodiments of this application is shown schematically.

[0026] Figure 5C The illustration shows a comparison of experimental results in different experimental scenarios according to embodiments of this application;

[0027] Figure 6 A table illustrating the experimental results of different positioning methods in different regions according to embodiments of this application is shown in the diagram.

[0028] Figure 7A A schematic diagram illustrating a first comparison of predicted values ​​and actual values ​​according to an embodiment of this application is shown.

[0029] Figure 7B A second comparison diagram of predicted values ​​and actual values ​​according to an embodiment of this application is illustrated schematically.

[0030] Figure 7C A third comparative diagram of predicted values ​​and actual values ​​according to an embodiment of this application is illustrated schematically.

[0031] Figure 8 A schematic diagram illustrating the structure of a dynamic scene indoor positioning device based on CSI data according to an embodiment of this application; and

[0032] Figure 9 A block diagram schematically illustrates an electronic device suitable for implementing a dynamic scene indoor positioning method based on CSI data, according to an embodiment of this application. Detailed Implementation

[0033] The embodiments of this application will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of this application. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of this application for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concepts of this application.

[0034] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0035] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0036] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0037] Significant progress has been made in the field of radio frequency-based indoor positioning. Researchers have established explicit models of the relationship between signal changes and personnel positions using advanced signal processing techniques. However, these explicit signal processing models often become difficult to solve when positioning tasks become more complex. Therefore, learning-based positioning methods based on deep neural networks have rapidly emerged, providing solutions for high-precision positioning tasks in many scenarios. While these methods are highly effective under specific conditions, they struggle to effectively handle dynamic interference, thus limiting their suitability for large-scale, highly complex applications.

[0038] Data-driven positioning methods typically require establishing a precise mapping between Channel State Information (CSI) and physical location, which is difficult to achieve in reality. In practical environments, dynamic factors such as human activity, object movement, and the opening and closing of doors and windows continuously alter the propagation path of wireless signals, leading to significant changes in signal characteristics. Modern CSI-based wireless sensing technologies utilize these changes to perceive the environmental state. Although previous wireless sensing research has usually modeled CSI from both static and dynamic component perspectives, this distinction has been largely ignored in the field of positioning research.

[0039] Currently, most existing indoor positioning research focuses on eliminating hardware biases by designing complex input features or optimizing network structure design to improve theoretical accuracy in specific environments. Therefore, existing methods inevitably couple dynamic signal components with static signal components. When the activity patterns of dynamic factors change, mismatches occur between the input features and the training data, leading to significant positioning errors.

[0040] This application provides a dynamic scene indoor positioning method based on CSI data, comprising: acquiring raw measurement data from N access points, the raw measurement data including raw channel state information and received signal strength, where N is a positive integer greater than or equal to 1; preprocessing the raw channel state information to obtain preprocessed channel state information; inputting the preprocessed channel state information into a temporal convolutional network for temporal feature extraction to generate hidden temporal features; based on the hidden temporal features, using a parallel one-dimensional convolutional network to simultaneously extract dynamic and static features; performing feature fusion on the static features based on a Transformer to obtain fused features; and inputting the fused features into a multilayer perceptron to obtain the predicted location. In this way, static and dynamic features can be separated from the channel state information. By using only static features for positioning, the interference noise of dynamic features can be accurately shielded to eliminate its misleading influence on the positioning results, ensuring accurate positioning.

[0041] Figure 1 The illustration shows an application scenario of the dynamic scene indoor positioning method, apparatus, and device based on CSI data according to embodiments of this application.

[0042] like Figure 1 As shown, application scenario 100 according to this embodiment may include target device 101, access point (AP) 102, network 103, and server 104. Target device 101 may communicate with the access point via wireless LAN, radio, and / or a wirelessly communicative radio. Although three access points are shown, more access points may be coupled to server 104.

[0043] Access point 102 may include access point 102a, access point 102b, and access point 103c. Access point 102 can be communicatively coupled to server 104 via network 103. All access points in access point 102 can share information about target device 101 to obtain channel state information on network 103. As will be understood, by directly or indirectly connecting access point 102 and server 104 and / or any of many other devices to network 103, access points 102 can communicate with each other, communicate with server 104, etc., thereby performing various functions for location, such as sending data, indications, information, etc. to each other, and / or receiving data, indications, information, etc. from each other.

[0044] As used herein, the terms “data,” “instruction,” “information,” and similar terms are used interchangeably to refer to data that can be transmitted, received, and / or stored according to embodiments of this application. Therefore, the use of any such terms should not be construed as limiting the spirit and scope of the embodiments of this application.

[0045] Access point 102 can be any type of network device configured to provide wireless access to target device 101 via a radio link, based on technologies such as radio frequency, infrared, or any of many different wireless network technologies, including wireless local area networks, global microwave access interoperability technology and / or wireless personal area network technology, Bluetooth, ultra-wideband, code division multiple access-based wireless cellular communication technology, high-speed packet data, universal terrestrial radio access network, Long Term Evolution (LTE), LTE Advanced, fifth-generation cellular systems, etc. It should be understood that the illustrated embodiments are non-limiting, and as will be readily understood by those skilled in the art, any number of various wireless devices and telecommunications systems can be employed.

[0046] Network 103 serves as the medium for providing a communication link between access point 102 and server 104. Network 103 may include various connection types, such as wired or wireless communication links or fiber optic cables, etc.

[0047] Server 104 can be any type of server, such as a Web (World Wide Web) or cloud server, application server, backend server, edge server, base station, personal computer, etc. Access point 102 and server 104 can be configured to support location services, and access point 102 can be configured to send location-related information of a target to server 104. For example, location-related information may include channel state information for communication with the target. Server 104 can be configured to collect location-related information of a target (such as target device 101) from access point 102, analyze and process the received information, and estimate the target's location. In some embodiments, the server may be installed in the same entity as the access point. In some embodiments, the server's functionality may be distributed across multiple entities, including the access point and the server.

[0048] From a positioning perspective, target device 101 can also be referred to as a terminal device, which means any terminal device capable of accessing a wireless communication network and receiving services therefrom. By way of example and not limitation, target device may refer to user equipment or other suitable devices, such as user stations, portable user stations, mobile stations, or access terminals, or any combination thereof. Target device may include, but is not limited to, portable computers, image capture devices such as digital cameras, gaming terminal devices, music storage and playback devices, mobile communication devices, mobile phones, cellular phones, smartphones, navigation devices, tablets, wearable devices, smartwatches, fitness bands, remote monitoring bands, personal digital assistants, vehicles, Internet of Things devices, sensor devices, etc., or any combination thereof.

[0049] It should be noted that the dynamic scene indoor positioning method based on CSI data provided in the embodiments of this application can generally be executed by server 104. Correspondingly, the dynamic scene indoor positioning device based on CSI data provided in the embodiments of this application can generally be located in server 104. The dynamic scene indoor positioning method based on CSI data provided in the embodiments of this application can also be executed by a server or server cluster that is different from server 104 but can communicate with access point 102 and / or server 104. Correspondingly, the dynamic scene indoor positioning device based on CSI data provided in the embodiments of this application can also be located in a server or server cluster that is different from server 104 but can communicate with access point 102 and / or server 104.

[0050] It should be understood that Figure 1 The number of target devices, access points, networks, and servers shown is merely illustrative. Depending on implementation needs, there can be any number of terminal devices, networks, and servers.

[0051] The following will be based on Figure 1 The described scene, through Figures 2-7CA dynamic scene indoor positioning method based on CSI data according to embodiments of this application will be described in detail.

[0052] Figure 2 A flowchart illustrating a dynamic scene indoor positioning method based on CSI data according to an embodiment of this application is shown.

[0053] like Figure 2 As shown, the dynamic scene indoor positioning method 200 based on CSI data in this embodiment includes operations S210 to S280.

[0054] In operation S210, in response to the target device's transmission signal being blocked by at least one person in the indoor environment, raw measurement data from N access points in the indoor environment is acquired. The raw measurement data includes raw channel state information, where N is a positive integer greater than or equal to 1.

[0055] In real-world applications, the propagation of wireless signals is influenced by a variety of complex factors, with human activity being a particularly prominent source of dynamic interference. Human movement, posture changes, and limb movements significantly alter the surrounding electromagnetic environment, causing signal fluctuations in multiple dimensions, including amplitude, phase, and time of arrival. Human movement patterns are diverse and time-varying; when activity patterns change, the statistical characteristics of the signal also change, leading to a significant deviation between the real-time acquired data and the data relied upon during model training. This discrepancy between measured and training data directly affects the accuracy of model predictions and increases positioning errors.

[0056] Access point (AP) measurement data can include one or more factors, such as signal strength (typically measured in decibels), received signal strength, signal-to-noise ratio (SNR), channel state information (CSI), etc. This data can be measured and recorded by wireless communication devices. CSI contains amplitude and phase information for each subcarrier, providing rich frequency domain information; the CSI on the subcarriers reflects the signal strength at different frequencies.

[0057] In the embodiments of this application, the raw measurement data from N access points includes raw channel state information and received signal strength, where N is a positive integer greater than or equal to 1. Received Signal Strength Indication (RSSI), also known as received signal strength, is an indicator used in wireless communication to represent the strength of the received signal, typically used to assess signal quality and the connection strength between the device and the network. Channel state information (CSI) is a set of parameters describing the channel characteristics in a wireless communication link, reflecting the impact of physical layer characteristics such as attenuation, fading, noise, and multipath effects experienced by the signal during transmission on signal quality.

[0058] In operation S220, at least one complex matrix of the original channel state information is reduced in dimension to obtain a flattened vector, thus obtaining the preprocessed original channel state information.

[0059] In the embodiments of this application, after obtaining the original channel state information, the channel state information can be preprocessed. The original channel state information represents a four-dimensional data matrix with dimensions N×T×M×K, where each access point provides T data packets, and each data packet represents a complex matrix with dimensions M×K, where T, M, and K are all positive integers greater than or equal to 1, M represents the number of antennas, and K represents the number of subcarriers. To unify the input format, each M×K two-dimensional submatrix in the third and fourth dimensions of the four-dimensional data matrix is ​​flattened to obtain a one-dimensional vector of length M×K; based on the flattening process, the four-dimensional data matrix can be reduced to a three-dimensional data matrix with dimensions N×T×MK, where MK=M×K.

[0060] In the embodiments of this application, the preprocessed original channel state information can be amplitude calibrated to obtain a calibrated data matrix, wherein the data structure of the preprocessed original channel state information is in matrix form; the channel frequency response power matrix is ​​obtained by squaring the absolute value of each element in the calibrated data matrix.

[0061] Automatic Gain Control (AGC) is a closed-loop feedback technique used in electronic circuits or signal processing systems. Its core function is to automatically adjust the signal amplification factor (gain) to maintain the stability of the output signal. In fingerprinting methods based on Channel State Information (CSI), a fundamental premise is that sampling points with similar spatial locations should also have highly similar CSI features (i.e., "fingerprints"). However, due to the presence of AGC, when environmental noise or other factors cause slight fluctuations in the received signal strength, AGC adjusts the gain in real time, thus scaling the overall extracted CSI amplitude. This process causes the CSI amplitude to no longer accurately reflect the physical characteristics of the channel, thereby weakening the stable correspondence between location and fingerprint. Therefore, in practical applications, the acquired raw CSI data is actually the result of nonlinear adjustment by AGC, and its amplitude information has been distorted. To restore the true characteristics of the channel response, these AGC-affected CSI amplitudes must be calibrated.

[0062] In embodiments of this application, to obtain the CSI amplitude unaffected by AGC, the Received Signal Strength (RSS) can be used to calibrate the CSI. Since the RSS is acquired before AGC, it reflects the original signal strength. In practical applications, AGC applies a uniform scaling factor s to all subcarriers, causing the CSI amplitude to be amplified or reduced overall. By utilizing the reference information provided by the RSS, this uniform scaling factor s can be estimated, and then used to remove the original CSI to recover the true CSI before AGC.

[0063] In the embodiments of this application, the amplitude calibration of the preprocessed original channel state information to obtain the calibrated data matrix specifically includes: calculating the scaling factor introduced by automatic gain control based on the received signal strength and the preprocessed original channel state information; dividing each element of the preprocessed original channel state information by the scaling factor to obtain the calibrated data matrix.

[0064] The scaling factor can be calculated using the following formula:

[0065]

[0066] Where s represents the scaling factor, and RSS represents the received signal strength. Represents any element of length in a three-dimensional data matrix. The i-th element in the one-dimensional vector; divide each element in the three-dimensional data matrix by the scaling factor to obtain the calibrated data matrix.

[0067] In the embodiments of this application, the absolute value of each element in the calibrated data matrix is ​​squared to obtain the channel frequency response power matrix, wherein the channel frequency response power matrix has a dimension of N×T×MK.

[0068] In the embodiments of this application, after obtaining the original channel state information, the original channel state information is first flattened to unify the input format and reduce the data dimension, thereby improving the efficiency of subsequent processing. Then, by calculating the scaling factor, the influence of automatic gain control on the amplitude of the channel state information can be removed, restoring its original signal before gain control, so as to eliminate its interference with subsequent positioning calculations. Afterwards, by taking the square of the absolute value of each element in the calibrated data matrix, the complex domain data that originally contained amplitude and phase information can be mapped to the non-negative real power domain. This data format conversion is essentially to refine the amplitude and phase mixed information into a scalar metric that directly represents the signal energy intensity, so that it can be directly used for positioning calculations.

[0069] In operation S230, the preprocessed channel state information is input into a temporal convolutional network to encode local temporal dependencies and generate hidden temporal features.

[0070] In the embodiments of this application, in order to capture temporal dynamic features and extract high-level semantic features, the preprocessed CSI data of each access point (AP) can be processed independently using Temporal Convolutional Networks (TCNs). Specifically, the TCN first captures local temporal dependencies in the sequence through its convolutional layers. Then, through layer-by-layer transformation, these basic dependencies are integrated and abstracted to finally generate hidden temporal features, i.e., intermediate features that can be used for subsequent processing, with a dimension of d. That is, after inputting the preprocessed channel state information into the temporal convolutional network for temporal feature extraction, the dimension of the matrix is ​​transformed from N×T×MK to N×T×d.

[0071] In operation S240, the hidden temporal features are processed using the first network branch to obtain dynamic features. These dynamic features represent the signals reflected from the transmitted signal after being blocked by at least one person and then transmitted to N access points.

[0072] In operation S250, the hidden temporal features are processed using the second network branch to obtain static features. Static features represent signals reflected from the transmitted signal after being blocked by an object other than at least one person, and then transmitted to N access points. The first and second network branches are trained according to different objective functions.

[0073] In the embodiments of this application, both the first network branch and the second network branch are generated based on training a one-dimensional convolutional network, wherein the first network branch is used to extract dynamic features and the second network branch is used to extract static features.

[0074] To decouple dynamic and static features, based on the hidden temporal features extracted by TCN, two one-dimensional convolutional network branches can be used to extract dynamic features. and static features Both of these feature types have dimensions of N×T×d.

[0075] It should be noted that the two one-dimensional convolutional network branches can be executed in parallel.

[0076] In operation S260, the static features are concatenated with the coordinate features of their corresponding access points to obtain the location-aware static features. The coordinate features represent the location of the corresponding access point within the indoor environment.

[0077] In the embodiments of this application, after feature decoupling is completed, stable static features suitable for localization can be separated from the original data containing mixed static and dynamic features. For static features, firstly, the static features can be averaged over time to obtain the average static feature for each access point. Since CSI is limited by hardware sampling accuracy, multipath fading, environmental interference, etc., it often contains a large amount of high-frequency noise. Therefore, averaging the static features over time can weaken the influence of instantaneous disturbances and random noise, making the feature curve smoother and thus improving the stability and consistency of the features. Then, since the Transformer module itself is insensitive to the order or spatial location of the input, relying only on the attention mechanism to capture the relationships between elements, for the average static features, the position encoding value of each access point coordinate can be calculated based on the position encoding function of the Transformer module, thus obtaining the coordinate features of the access point. Each average static feature is then concatenated with its corresponding position encoding value. This position encoding explicitly injects the position information of the access point into the average static feature. The concatenation formula is expressed as:

[0078]

[0079] in, This represents the average static feature (i.e., position-aware static feature) after concatenating the positional encoding values. This represents the s-th average static feature. This represents the position encoding function of the Transformer module. These are the coordinates of the corresponding access point.

[0080] In the embodiments of this application, after concatenating the static features with the coordinate features of their corresponding access points to obtain the location-aware static features, the method includes: adding a classification token as a prefix to the location-aware static features, wherein the classification token is used to identify feature category information at the front end of the location-aware static feature sequence.

[0081] Classification tokens, through self-attention interactions with all other locations, can aggregate global information to form a semantically rich "summary representation," where the classification token addition formula is expressed as:

[0082]

[0083] in, This indicates the feature after adding the category token. A category token representing a location-coded value.

[0084] In the embodiments of this application, when N is greater than or equal to 2, multiple static features are concatenated with multiple coordinate features, wherein each static feature corresponds to an access point, and the multiple coordinate features represent multiple access points distributed in multiple different locations in the indoor environment.

[0085] When operating the S270, the Transformer module is used to process the position-aware static features to obtain position-aware fused features.

[0086] In the embodiments of this application, the method of obtaining location-aware fusion features by performing feature processing on location-aware static features based on Transformer modules specifically includes: inputting location-aware static features into multiple Transformer modules cascaded in sequence for feature transformation processing, wherein each Transformer module performs spatial location information interaction and fusion on the location-aware static features through a self-attention mechanism; and generating location fusion features based on the features output by multiple Transformer modules.

[0087] In the embodiments of this application, after calculating the average value, adding location encoding, and adding classification tokens to the static features, they can be passed through the Transformer module for feature fusion across access points. The location-aware fused features obtained by fusing location-aware static features based on the Transformer module can be achieved using the following formula:

[0088]

[0089] in, Indicates initial features, Indicates the first The input feature representation of the layer, This indicates application layer normalization. This indicates that multi-head self-attention computation is performed on the input sequence. Indicates intermediate features, This represents a feedforward neural network; Indicates the first Layer output features.

[0090] It should be noted that for feature processing, multiple Transformer modules can be cascaded in sequence to perform feature transformation. Each Transformer module uses a self-attention mechanism to interact and aggregate positional information in the spatial dimension, capturing long-range dependencies while preserving local details. As the layers progress from shallow to deep, the model can gradually abstract low-level texture and edge information into high-level semantic features, thereby improving feature discriminability and robustness, and building more general and hierarchical feature representations for downstream tasks.

[0091] Single-AP CSI (Channel Signaling System) can only capture the one-way channel response from the target to that AP, failing to comprehensively depict spatial characteristics. Furthermore, due to limitations in signal coverage, CSI is highly sensitive to environmental changes (such as human movement, furniture obstruction, and multipath reflection). Single-AP CSI is prone to feature distortion (such as sudden amplitude drops and phase jumps) due to local obstruction (e.g., the target being blocked by a wall / person) or multipath superposition (e.g., reflected signals masking direct signals), leading to sensing failure. In the embodiments of this application, by utilizing Transformer for cross-access point feature fusion, observation data from access points at different locations can be integrated to obtain more comprehensive features, thereby eliminating the spatial blind spots of a single AP.

[0092] In operation S280, the location-aware fused features are input into the multilayer perceptron to obtain the predicted location.

[0093] A multilayer perceptron (MLP) is a feedforward artificial neural network model that maps an input dataset to a single output dataset.

[0094] In embodiments of this application, the fused features (i.e., location-aware fused features) can be input into a multilayer perceptron to predict the user's location coordinates. .

[0095]

[0096] in, This indicates that a multilayer perceptron is used for feature processing. Indicates the first Data features of classification tokens extracted by layer network (i.e., location-aware fusion features).

[0097] In the embodiments of this application, in order to achieve more accurate and efficient dynamic scene indoor positioning, a feature decoupling strategy is adopted, which decouples the channel state information into two dimensions: static features and dynamic features. By using only the static features obtained from the decoupling separation for positioning, the interference noise generated by the dynamic features can be accurately shielded to eliminate its misleading influence on the positioning results and ensure accurate positioning.

[0098] Figure 3 A flowchart illustrating the setting of a loss function according to an embodiment of this application is shown.

[0099] like Figure 3 As shown, the method 300 of this embodiment includes operations S310 to S340.

[0100] In operation S310, the features obtained from the execution of the first and second network branches are input into a pre-constructed set of first loss functions. This set of first loss functions includes a similarity loss function, an orthogonal loss function, and a reconstruction loss function.

[0101] In embodiments of this application, the first set of loss functions includes a similarity loss function, an orthogonal loss function, and a reconstruction loss function.

[0102] The similarity loss function is used to minimize the differences between static features at consecutive time points, and its calculation formula is expressed as:

[0103]

[0104] in, Represents the similarity loss function. Indicates the number of access points. This indicates the number of data packets provided by each access point. This represents the static characteristics of the nth access point at time t. Indicates that the nth access point is in The static characteristics of time, This represents the square of the L2 norm.

[0105] In the embodiments of this application, in order to constrain the consistency of static features over time, the difference between frames is minimized by a similarity loss function to penalize changes over time. This can make the features of adjacent time steps as similar as possible, thereby ensuring the continuity and consistency of features in the time dimension and preventing the model from learning unstable or jumping representations.

[0106] The orthogonal loss function is used to minimize the correlation between static and dynamic features, and its calculation formula is expressed as:

[0107]

[0108] in, Represents the orthogonal loss function. Indicates the number of access points. This indicates the number of data packets provided by each access point. This represents the static characteristics of the nth access point at time t. This represents the dynamic characteristics of the nth access point at time t. This represents the L2 norm.

[0109] In the embodiments of this application, in order to ensure that the static features and dynamic features are independent of each other, their inner product is constrained by an orthogonal loss function, which can make them as separate as possible in the feature space and enhance the discriminability between the two types of features.

[0110] The reconstruction loss function is used to minimize the difference between the sum of the reconstructed static and dynamic signals and the input signal. Its calculation formula is as follows:

[0111]

[0112] in, Represents the reconstruction loss function. Indicates the number of access points. This indicates the number of data packets provided by each access point. This represents the original hidden features of the nth access point at time t. This represents the reconstructed static signal component (i.e., static signal) extracted by the nth access point at time t. This represents the reconstructed dynamic signal component (i.e., dynamic signal) extracted by the nth access point at time t. This represents the square of the L2 norm.

[0113] In the embodiments of this application, in order to reconstruct the original hidden features by the sum of the decoded static features and dynamic features, a reconstruction loss function is used to constrain the process so that the combined split features can restore the original information as much as possible, avoid information loss or distortion during the splitting process, and maintain information integrity.

[0114] In operation S320, the output of the multilayer perceptron is input into a pre-constructed set of second loss functions. This set of second loss functions includes a prediction location loss function.

[0115] In embodiments of this application, the second set of loss functions includes a prediction location loss function.

[0116] The prediction location loss function is used to minimize the error between the predicted location and the true location, and its calculation formula is expressed as:

[0117]

[0118] in, This represents the loss function for predicting location. Indicates the predicted location. Indicates the actual location. This represents the square of the L2 norm.

[0119] In the embodiments of this application, by creating a prediction location loss function, the difference between the predicted location and the actual location can be accurately measured. By minimizing the value of this loss function, the model's prediction results are directly driven to approximate the actual situation, thereby improving the accuracy of the predicted location.

[0120] In operation S330, all loss function values ​​in the first loss function set and the second loss function set are weighted and fused to obtain the total loss function value.

[0121] In the embodiments of this application, the total loss function value can be obtained by weighted fusion of the loss function values ​​of the similarity loss function, orthogonality loss function, and reconstruction loss function of the first loss function set with the loss function value of the predicted position loss function of the second loss function set. The formula for calculating the total loss function can be expressed as follows:

[0122]

[0123] in, , , , These represent the weight parameters of the similarity loss function, orthogonality loss function, reconstruction loss function, and prediction location loss function, respectively.

[0124] It should be noted that the specific value of each weight parameter can be determined according to actual needs, and no restrictions are imposed here.

[0125] In operation S340, based on the total loss function value, the parameters of the temporal convolutional network, the first network branch, the second network branch, the Transformer module, and the multilayer perceptron are updated through gradient backpropagation.

[0126] In the embodiments of this application, the total loss function value can be calculated based on this total loss function during the training process. By applying gradient backpropagation to the entire network (temporal convolutional network, one-dimensional convolutional network (i.e., the first network branch and the second network branch), Transformer module, and multilayer perceptron), end-to-end joint optimization is achieved.

[0127] In the embodiments of this application, by setting a loss function, the static features and dynamic features maintain consistent physical semantics across all modules, providing a clear direction for model parameter updates, thereby effectively improving the accuracy and stability of subsequent localization tasks.

[0128] It should be noted that, in the embodiments of this application, each loop executes the temporal convolutional network, the one-dimensional convolutional network, the Transformer module, and the multilayer perceptron in sequence. The number of loops can be predefined by the user; after the specified number of loops is met, the process terminates and the final predicted position is output.

[0129] Optionally, the absolute difference between the most recent predicted positions can be continuously monitored. Once the difference stabilizes (remains unchanged) or falls below the set absolute difference threshold, the judgment result has converged, and the loop automatically stops.

[0130] For example, a real mobile terminal can be used as the positioning target, and an access point (AP) can be deployed as the data acquisition terminal. Each AP is equipped with two antennas, and the carrier frequency of the WiFi signal is 5GHz. During positioning, firstly, the CSI signals measured by multiple APs are preprocessed to obtain input data; secondly, static data components (i.e., static features) are decoupled from the input data; finally, positioning is performed based on the static data components, and the coordinate position is output.

[0131] Figure 4 Another flowchart illustrating a dynamic scene indoor positioning method based on CSI data according to an embodiment of this application is shown.

[0132] like Figure 4 As shown, the dynamic scene indoor positioning method based on CSI data mainly consists of three steps: step 1 to step 3.

[0133] Step 1: Data Preprocessing. The obtained raw channel state information (i.e., the raw CSI information shown in the figure) is preprocessed to obtain preprocessed channel state information (i.e., the preprocessed CSI information shown in the figure).

[0134] Step 2: Feature Separation. The preprocessed channel state information is input into a temporal convolutional network for temporal feature extraction, generating hidden temporal features. Then, based on these hidden temporal features, two parallel one-dimensional convolutional networks are used to simultaneously extract dynamic and static features. Each one-dimensional convolutional network includes one-dimensional convolution and one-dimensional deconvolution. Intermediate dynamic and static features are generated through one-dimensional convolution and then input into their respective one-dimensional deconvolutions. Based on the features obtained from the one-dimensional convolutional network process, the corresponding loss function values ​​can be calculated using similarity loss function, orthogonality loss function, and reconstruction loss function, respectively.

[0135] Step 3: Localization. First, the static features are averaged over time to obtain average static features. These average static features are then concatenated with the location-encoded values ​​(i.e., the AP coordinates shown in the diagram) to obtain location-aware static features. These are then input into the Transformer module for feature fusion. The fusion operation uncovers hidden location features, which are the location-aware fused features that can be used for subsequent data processing. Next, the obtained hidden location features are input into a multilayer perceptron to obtain the predicted location. Based on the output of the multilayer perceptron, the predicted location loss function value can be calculated using the predicted location loss function.

[0136] Then, based on the similarity loss function value, orthogonality loss function value, and reconstruction loss function value obtained in step 2, and the prediction location loss function value obtained in step 3, the total loss function value is calculated using the total loss function. Based on the total loss function value, the parameters of the temporal convolutional network, the one-dimensional convolutional network, the Transformer module, and the multilayer perceptron are updated through gradient backpropagation.

[0137] Figure 5A This is a table illustrating the experimental scenarios of an embodiment of this application. The experimental scenarios used in the embodiments of this application are all interference scenarios, used to simulate positioning accuracy under different interference conditions. Figure 5A As shown, interference scenario number 1 is set to no interference, that is, no interference factor is set; interference scenarios numbered 2 to 6 are set to a single interference factor, while interference scenarios numbered 7 to 10 are composed of multiple single interference factors to simulate a more complex mixed interference environment.

[0138] Figure 5B A tabular diagram illustrating the experimental methods according to embodiments of this application is shown schematically. Figure 5B As shown, five other positioning methods (experimental methods 1 to 5) were compared with the dynamic scene indoor positioning method based on CSI data of the embodiment of this application (experimental method 6).

[0139] Figure 5C These are different experimental scenarios based on embodiments of this application (i.e.) Figure 5A A schematic diagram comparing experimental results in the experimental scenario shown. Figure 5C As shown, the vertical axis represents the magnitude of the positioning error in meters; the horizontal axis represents the sequence number of the interference scene, which was used sequentially. Figure 5B The experimental methods shown were compared, and the evaluation metrics included the mean and 90th percentile (P90). Specifically, for each experimental method, the mean and 90th percentile of the positioning error were calculated for a given experimental scenario. Figure 5C As can be seen from the embodiments, the position error predicted by the method of this application is significantly smaller than that of other positioning methods.

[0140] Figure 6 A comparative table illustrating experimental results of different positioning methods in different regions according to embodiments of this application is shown. Figure 6 As shown, the experimental method used is Figure 5B The experimental method shown has numerical values ​​representing the magnitude of error. 2A, 2B, 3A, 3B3C, 4A, 4B4C, 5A, and 5B5C represent different experimental regions. The mean is the average of the results from different regions. Figure 6It can be seen that the position error predicted by the method of the embodiment of this application (i.e., experimental method 6 shown in the figure) is significantly smaller than that of other positioning methods.

[0141] Figure 7A A schematic diagram illustrating a first comparison of predicted values ​​and actual values ​​according to an embodiment of this application is shown. Figure 7B A second comparison diagram of predicted values ​​and actual values ​​according to an embodiment of this application is illustrated schematically. Figure 7C A third comparative diagram of predicted values ​​and actual values ​​according to an embodiment of this application is illustrated. Figure 7A , Figure 7B , Figure 7C The symbol "X" in the text represents location prediction, which means the predicted target location, i.e., the predicted value. Figure 7A , Figure 7B , Figure 7C The symbol in "" indicates a location label, representing the actual target location, i.e., the actual value; Figure 7A , Figure 7B , Figure 7C The letter AP in the name indicates an access point. Figure 7A , Figure 7B , Figure 7C The diagrams show the comparison between predicted and actual values ​​for three different devices in three different regions.

[0142] Based on the above-mentioned dynamic scene indoor positioning method based on CSI data, this application also provides a dynamic scene indoor positioning device based on CSI data. The following will combine... Figure 8 The device is described in detail.

[0143] Figure 8 A schematic diagram illustrating the structure of a dynamic scene indoor positioning device based on CSI data according to an embodiment of this application is shown.

[0144] like Figure 8 As shown, the dynamic scene indoor positioning device 800 based on CSI data in this embodiment includes a raw measurement data acquisition module 810, a channel state information preprocessing module 820, a time feature generation module 830, a dynamic feature extraction module 840, a static feature extraction module 850, a location-aware static feature generation module 860, a location-aware fusion feature acquisition module 870, and a predicted location determination module 880.

[0145] The raw measurement data acquisition module 810 is used to acquire raw measurement data from N access points in the indoor environment in response to the target device's transmission signal being blocked by at least one person. The raw measurement data includes raw channel state information, where N is a positive integer greater than or equal to 1. In one embodiment, the raw measurement data acquisition module 810 can be used to perform the operation S210 described above, which will not be repeated here.

[0146] The channel state information preprocessing module 820 is used to reduce the dimensionality of at least one complex matrix of the original channel state information to obtain a flattened vector, thereby obtaining the preprocessed original channel state information. In one embodiment, the channel state information preprocessing module 820 can be used to perform the operation S220 described above, which will not be repeated here.

[0147] The temporal feature generation module 830 is used to input the preprocessed channel state information into a temporal convolutional network to encode local temporal dependencies and generate hidden temporal features. In one embodiment, the temporal feature generation module 830 can be used to perform the operation S230 described above, which will not be repeated here.

[0148] The dynamic feature extraction module 840 is used to process the hidden temporal features using the first network branch to obtain dynamic features, which represent the signals reflected to N access points after being blocked by at least one person. In one embodiment, the dynamic feature extraction module 840 can be used to perform the operation S240 described above, which will not be repeated here.

[0149] The static feature extraction module 850 is used to process the hidden temporal features using the second network branch to obtain static features, which represent the signals reflected to N access points after being blocked by objects other than at least one person. In one embodiment, the static feature extraction module 850 can be used to perform the operation S250 described above, which will not be repeated here.

[0150] The location-aware static feature generation module 860 is used to concatenate static features with the coordinate features of their corresponding access points to obtain location-aware static features, where the coordinate features characterize the location of the corresponding access point in the indoor environment. In one embodiment, the location-aware static feature generation module 860 can be used to perform the operation S260 described above, which will not be repeated here.

[0151] The location-aware fusion feature acquisition module 870 is used to perform feature processing on the location-aware static features based on the Transformer module to obtain location-aware fusion features. In one embodiment, the location-aware fusion feature acquisition module 870 can be used to perform the operation S270 described above, which will not be repeated here.

[0152] The predicted location determination module 880 is used to input the location-aware fusion features into a multilayer perceptron to obtain the predicted location. In one embodiment, the predicted location determination module 880 can be used to perform the operation S280 described above, which will not be repeated here.

[0153] According to an embodiment of this application, the dynamic scene indoor positioning device 800 based on CSI data further includes: a first loss function set module, a second loss function set module, a total loss function module, and a parameter adjustment module.

[0154] The first loss function set module is used to input the features obtained from the execution of the first network branch and the second network branch into a pre-constructed first loss function set, which includes a similarity loss function, an orthogonal loss function, and a reconstruction loss function.

[0155] The second loss function set module is used to input the output of the multilayer perceptron into a pre-constructed second loss function set, which includes a prediction location loss function.

[0156] The total loss function module is used to weight and fuse all loss function values ​​in the first loss function set and the second loss function set to obtain the total loss function value.

[0157] The parameter adjustment module is used to update the parameters of the temporal convolutional network, the first network branch, the second network branch, the Transformer module, and the multilayer perceptron based on the total loss function value through gradient backpropagation.

[0158] According to an embodiment of this application, the dynamic scene indoor positioning device 800 based on CSI data further includes: a channel frequency response power matrix generation module.

[0159] The channel frequency response power matrix generation module is used to perform amplitude calibration on the preprocessed raw channel state information to obtain a calibrated data matrix. The data structure of the preprocessed raw channel state information is in matrix form. The channel frequency response power matrix is ​​obtained by squaring the absolute value of each element in the calibrated data matrix.

[0160] According to embodiments of this application, any and multiple modules among the raw measurement data acquisition module 810, channel state information preprocessing module 820, time feature generation module 830, dynamic feature extraction module 840, static feature extraction module 850, location-aware static feature generation module 860, location-aware fusion feature acquisition module 870, and predicted location determination module 880 can be combined into one module, or any one of these modules can be split into multiple modules. Alternatively, at least some of the functions of one or more of these modules can be combined with at least some of the functions of other modules and implemented in one module. According to embodiments of this application, at least one of the following modules can be implemented, at least partially, as hardware circuits: the raw measurement data acquisition module 810, the channel state information preprocessing module 820, the time feature generation module 830, the dynamic feature extraction module 840, the static feature extraction module 850, the position-aware static feature generation module 860, the position-aware fusion feature acquisition module 870, and the predicted position determination module 880. These can be implemented, for example, as hardware or firmware circuits such as field-programmable gate arrays (FPGAs), programmable logic arrays (PLAs), systems-on-a-chip, systems-on-a-substrate, systems-on-package, application-specific integrated circuits (ASICs), or any other reasonable means of integrating or packaging circuits. Alternatively, they can be implemented in any one of the three methods of software, hardware, and firmware, or in a suitable combination of any of these methods. Alternatively, at least one of the raw measurement data acquisition module 810, channel state information preprocessing module 820, time feature generation module 830, dynamic feature extraction module 840, static feature extraction module 850, location-aware static feature generation module 860, location-aware fusion feature acquisition module 870, and predicted location determination module 880 can be at least partially implemented as a computer program module, which can perform corresponding functions when the computer program module is run.

[0161] Figure 9 A block diagram schematically illustrates an electronic device suitable for implementing a dynamic scene indoor positioning method based on CSI data, according to an embodiment of this application.

[0162] like Figure 9As shown, an electronic device 900 according to an embodiment of this application includes a processor 901, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 902 or a program loaded from a storage portion 908 into a random access memory (RAM) 903. The processor 901 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 901 may also include onboard memory for caching purposes. The processor 901 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of this application.

[0163] RAM 903 stores various programs and data required for the operation of electronic device 900. Processor 901, ROM 902, and RAM 903 are interconnected via bus 904. Processor 901 executes various operations of the method flow according to embodiments of this application by executing programs in ROM 902 and / or RAM 903. It should be noted that the programs may also be stored in one or more memories other than ROM 902 and RAM 903. Processor 901 may also execute various operations of the method flow according to embodiments of this application by executing programs stored in said one or more memories.

[0164] According to embodiments of this application, the electronic device 900 may further include an input / output (I / O) interface 905, which is also connected to a bus 904. The electronic device 900 may also include one or more of the following components connected to the input / output (I / O) interface 905: an input section 906 including a keyboard, mouse, etc.; an output section 907 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 908 including a hard disk, etc.; and a communication section 909 including a network interface card such as a LAN card, modem, etc. The communication section 909 performs communication processing via a network such as the Internet. A drive 910 is also connected to the input / output (I / O) interface 905 as needed. A removable medium 911, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 910 as needed so that computer programs read from it can be installed into the storage section 908 as needed.

[0165] This application also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of this application.

[0166] According to embodiments of this application, the computer-readable storage medium can be a non-volatile computer-readable storage medium, such as including but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this application, the computer-readable storage medium may include ROM 902 and / or RAM 903 and / or one or more memories other than ROM 902 and RAM 903 described above.

[0167] Embodiments of this application also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code enables the computer system to implement the CSI data-based dynamic scene indoor positioning method provided by embodiments of this application.

[0168] When the computer program is executed by the processor 901, it performs the functions defined in the system / apparatus of the embodiments of this application. According to the embodiments of this application, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0169] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and downloaded and installed via the communication section 909, and / or installed from a removable medium 911. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0170] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 909, and / or installed from the removable medium 911. When the computer program is executed by the processor 901, it performs the functions defined in the system of the embodiments of this application. According to the embodiments of this application, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0171] According to embodiments of this application, program code for executing the computer programs provided in embodiments of this application can be written using any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0172] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0173] Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application.

Claims

1. A dynamic scene indoor positioning method based on CSI data, characterized in that, The method includes: In response to the target device's transmission signal being blocked by at least one person in the indoor environment, raw measurement data from N access points in the indoor environment is acquired. The raw measurement data includes raw channel state information, where N is a positive integer greater than or equal to 1. At least one complex matrix of the original channel state information is reduced in dimensionality to obtain a flattened vector, thus obtaining the preprocessed original channel state information. The preprocessed channel state information is input into a temporal convolutional network to encode local temporal dependencies and generate hidden temporal features. The hidden temporal features are processed using the first network branch to obtain dynamic features, which represent the signals of the transmitted signal reflected to the N access points after being blocked by at least one person. The hidden temporal features are processed using a second network branch to obtain static features, which represent the signals of the transmitted signal reflected to the N access points after being blocked by objects other than the at least one person. The static feature is concatenated with the coordinate feature of its corresponding access point to obtain the location-aware static feature, wherein the coordinate feature represents the position of the corresponding access point in the indoor environment. The location-aware static features are processed using the Transformer module to obtain location-aware fusion features. The location-aware fusion features are input into a multilayer perceptron to obtain the predicted location.

2. The method according to claim 1, characterized in that, The step of concatenating the static features with the coordinate features of their corresponding access points to obtain location-aware static features includes: When N is greater than or equal to 2, the multiple static features are concatenated with the multiple coordinate features, wherein each static feature corresponds to one access point, and the multiple coordinate features represent the multiple access points distributed in multiple different locations in the indoor environment.

3. The method according to claim 1, characterized in that, The method further includes: The features obtained from the execution of the first network branch and the second network branch are input into a pre-constructed set of first loss functions, wherein the set of first loss functions includes a similarity loss function, an orthogonal loss function, and a reconstruction loss function; The output of the multilayer perceptron is input into a pre-constructed set of second loss functions, wherein the set of second loss functions includes a prediction location loss function; The total loss function value is obtained by weighted and fused together all the loss function values ​​in the first set of loss functions and the second set of loss functions. Based on the total loss function value, the parameters of the temporal convolutional network, the first network branch, the second network branch, the Transformer module, and the multilayer perceptron are updated through gradient backpropagation.

4. The method according to claim 3, characterized in that, The similarity loss function is used to minimize the difference between the static features at consecutive time points; the orthogonal loss function is used to minimize the correlation between the static features and the dynamic features; the reconstruction loss function is used to minimize the difference between the sum of the reconstructed static and dynamic signals and the input signal; and the prediction position loss function is used to minimize the error between the predicted position and the true position.

5. The method according to claim 1, characterized in that, The method further includes: The preprocessed raw channel state information is subjected to amplitude calibration to obtain a calibrated data matrix, wherein the data structure of the preprocessed raw channel state information is in matrix form; The channel frequency response power matrix is ​​obtained by squaring the absolute value of each element in the calibrated data matrix.

6. The method according to claim 5, characterized in that, The step of performing amplitude calibration on the preprocessed raw channel state information to obtain a calibrated data matrix includes: Based on the received signal strength and the preprocessed original channel state information, the scaling factor introduced by automatic gain control is calculated. Divide each element of the preprocessed raw channel state information by the scaling factor to obtain the calibrated data matrix.

7. The method according to claim 1, characterized in that, After concatenating the static features with the coordinate features of their corresponding access points to obtain the location-aware static features, the following steps are included: A classification token is added as a prefix to the location-aware static feature, wherein the classification token is used to identify feature category information at the front end of the location-aware static feature sequence.

8. The method according to claim 1, characterized in that, The step of performing feature processing on the location-aware static features based on the Transformer module to obtain location-aware fused features includes: The location-aware static features are input into multiple cascaded Transformer modules for feature transformation processing. Each Transformer module interacts and fuses spatial location information with the location-aware static features through a self-attention mechanism. Based on the features output by the multiple Transformer modules, location fusion features are generated.

9. A dynamic scene indoor positioning device based on CSI data, characterized in that, The device includes: The raw measurement data acquisition module is used to acquire raw measurement data from N access points in the indoor environment in response to the target device's transmission signal being blocked by at least one person. The raw measurement data includes raw channel state information, where N is a positive integer greater than or equal to 1. The channel state information preprocessing module is used to perform dimensionality reduction processing on at least one complex matrix of the original channel state information to obtain a flattened vector, thereby obtaining the preprocessed original channel state information. The temporal feature generation module is used to input the preprocessed channel state information into a temporal convolutional network to encode local temporal dependencies and generate hidden temporal features. The dynamic feature extraction module is used to process the hidden temporal features using a first network branch to obtain dynamic features, wherein the dynamic features represent the signals of the transmitted signal reflected to the N access points after being blocked by at least one person. A static feature extraction module is used to process the hidden temporal features using a second network branch to obtain static features. The static features represent the signals reflected to the N access points after the transmitted signal is blocked by an object other than the at least one person. The first network branch and the second network branch are trained according to different objective functions. A location-aware static feature generation module is used to concatenate the static feature with the coordinate features of its corresponding access point to obtain a location-aware static feature, wherein the coordinate features characterize the position of the corresponding access point in the indoor environment. A location-aware fusion feature acquisition module is used to perform feature processing on the location-aware static features based on the Transformer module to obtain location-aware fusion features; and The predicted location determination module is used to input the location-aware fusion features into a multilayer perceptron to obtain the predicted location.

10. An electronic device, comprising: One or more processors; Memory, used to store one or more computer programs. The characteristic feature is that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 8.

11. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 8.

12. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 8.

Images