Passive positioning method, device and equipment based on WIFI wireless signal and medium

Abstract

The application discloses a passive positioning method and device based on WIFI wireless signals, equipment and a medium, relates to the technical field of wireless communication and positioning, and comprises the following steps: constructing an initial signal sample set; preprocessing the initial signal sample set to generate a standardized signal sequence; extracting fine-grained features related to the target position to construct a feature space; inputting the feature space into a positioning model to output the position estimation result of the target; iteratively optimizing feature extraction; outputting the final positioning result; generating a time alignment parameter matrix by aligning the timestamps of the initial complex matrix, and passing the time alignment parameter matrix for phase compensation, which overcomes the inconsistency of phase information in the traditional method; constructing a dynamic feature descriptor by performing feature decomposition on the weighted covariance matrix, which realizes the joint representation of the target motion state and the spatial position, enables the positioning model to accurately distinguish targets with different positions and motion states, and significantly improves the passive positioning accuracy.

Description

Technical Field

[0001] This invention relates to the field of wireless communication and positioning technology, specifically to a passive positioning method, apparatus, device, and medium based on WIFI wireless signals. Background Technology

[0002] WiFi wireless signals are affected by the environment during propagation, and their channel state information reflects the amplitude attenuation and phase changes along the signal propagation path. Passive positioning technology based on WiFi signals utilizes the interference of human bodies or targets on signal propagation, and achieves positioning without wearable devices by analyzing the spatiotemporal changes of channel state information. In a typical multi-receiver passive positioning system, signals collected by different receivers need to be aligned based on a unified time reference to eliminate phase inconsistencies caused by differences in signal arrival times. Simultaneously, target movement causes a micro-Doppler effect in the signal, manifested as dynamic phase changes in the channel state information, a feature closely related to the target's motion state and spatial location. Existing technologies typically achieve synchronization of signals from multiple receivers through timestamp alignment and use dimensionality reduction methods such as principal component analysis to extract features related to the target's location, subsequently employing machine learning models for location estimation.

[0003] However, in the existing technology, the time alignment parameters generated by the time alignment operation are only used for signal synchronization and are not fully utilized in the subsequent feature extraction process. As a result, the phase offset information introduced by the difference in signal propagation path and the micro-Doppler phase features generated by the target motion cannot be effectively fused. This results in insufficient representation of the coupling relationship between the target motion state and spatial position in the feature space, which limits the positioning accuracy in complex dynamic scenes. Summary of the Invention

[0004] The purpose of this invention is to provide a passive positioning method, device, equipment, and medium based on WIFI wireless signals, in order to solve the problem that in the prior art, the phase offset information introduced by the difference in signal propagation path and the micro-Doppler phase features generated by the target motion cannot be effectively fused, resulting in insufficient characterization of the coupling relationship between the target motion state and spatial position in the feature space, which limits the positioning accuracy in complex dynamic scenes.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a passive positioning method based on WIFI wireless signals, comprising the following steps:

[0006] S1: Collect WiFi channel status information in the environment and construct an initial signal sample set;

[0007] S2: Preprocess the initial signal sample set to remove outliers and noise, and generate a standardized signal sequence;

[0008] S3: Based on the standardized signal sequence, extract fine-grained features related to the target location and construct a feature space;

[0009] S4: Input the feature space into the pre-trained localization model and output the target's position estimation result;

[0010] S5: Based on the location estimation results, dynamically update the decision boundary of the localization model and feed it back to step S3 to iteratively optimize feature extraction;

[0011] S6: Output the final localization result.

[0012] Furthermore, step S1 further includes:

[0013] S11: Simultaneously collect WiFi data packets from the same transmitter at multiple receiving ends and record the original channel state information of each data packet;

[0014] S12: Extract the amplitude and phase information of each data packet to form an initial complex matrix;

[0015] S13: Timestamp-align the initial complex matrix to construct the initial signal sample set;

[0016] S14: Perform preliminary screening on the initial signal sample set and remove samples with a signal-to-noise ratio lower than a preset threshold.

[0017] Furthermore, step S3 further includes:

[0018] S31: Perform wavelet transform on the standardized signal sequence to extract multi-scale energy distribution features;

[0019] S32: Principal component analysis is used to reduce the dimensionality of the multi-scale energy distribution features to generate principal component feature vectors;

[0020] S33: Based on the principal component feature vector, construct fine-grained features that combine the time domain and frequency domain;

[0021] S34: Normalize the fine-grained features to form the feature space.

[0022] Furthermore, in step S13, when performing timestamp alignment on the initial complex matrix, a time alignment parameter matrix is ​​generated:

[0023]

[0024] in, For the number of receivers, For the number of data packets, This represents the offset relative to the reference timestamp when the nth receiver receives the i-th data packet;

[0025] Wherein, the time alignment parameter matrix Passed to step S32;

[0026] In step S32, during the dimensionality reduction process using principal component analysis, the time alignment parameter matrix is ​​used as a reference. Phase compensation is performed on the multi-scale energy distribution features so that the extracted principal component feature vectors contain phase shift information introduced by the differences in signal propagation paths;

[0027] The phase compensation is performed according to the following formula:

[0028]

[0029] in, These are the complex coefficients corresponding to the multi-scale energy distribution characteristics. The center frequency of the WIFI signal. For the elements in the time alignment parameter matrix, These are the complex coefficients after phase compensation. The natural constant is 2.718.

[0030] Furthermore, the time alignment parameter matrix generated in step S13 The elements in Specifically, it is used to construct the phase compensation matrix in step S32. :

[0031]

[0032] Step S32 is based on the phase compensation matrix. The multi-scale energy distribution features are rotated in the complex domain to ensure that the principal component feature vectors contain micro-Doppler phase features caused by the target motion.

[0033] The elements of the rotated-corrected feature matrix are represented as follows:

[0034]

[0035] in, Represents the complex coefficients after rotation correction. Let be the element in the nth row and ith column of the multi-scale energy distribution feature matrix after wavelet transform.

[0036] Furthermore, the time alignment parameter matrix generated in step S13 The elements in This is further used in step S32 to determine the weighting coefficients of the covariance matrix of the principal component analysis;

[0037] First, calculate the time alignment parameter matrix. Along the gradient of the time dimension, for the nth receiver, the time offset difference between adjacent data packets is:

[0038]

[0039] Constructing the gradient matrix :

[0040]

[0041] Construct a weighted coefficient matrix based on the gradient matrix. :

[0042]

[0043] Among them, diagonal elements Calculate according to the following formula:

[0044]

[0045] in, Preset scale parameters;

[0046] Step S32 is based on the weighting coefficient matrix. Enhance the phase component related to the target motion state and suppress the phase component corresponding to static multipath interference, thereby making the phase change feature caused by the target motion in the principal component feature vector significantly prominent;

[0047] The covariance matrix is ​​calculated using the following weighted formula:

[0048]

[0049] in, The diagonal matrix formed by taking the square roots of each diagonal element of the weighted coefficient matrix is ​​as follows:

[0050]

[0051] Where, (x) T This indicates that the matrix x is transposed.

[0052] Furthermore, the weighted covariance matrix generated in step S32 Passed to step S33;

[0053] In step S33, when constructing the fine-grained features jointly in the time and frequency domains, the weighted covariance matrix is... Perform feature decomposition, extract the principal feature values ​​and their corresponding principal feature vectors, and construct dynamic feature descriptors;

[0054] The eigenvalue decomposition is performed according to the following formula:

[0055]

[0056] in, For the first 1 eigenvalue, Let be the corresponding eigenvectors, and let the eigenvalues ​​be arranged in descending order. ;

[0057] Before selection There are ___ principal eigenvalues ​​and their corresponding principal eigenvectors, where Construct the main feature value vector:

[0058]

[0059] Construct the principal feature vector matrix:

[0060]

[0061] in, For the first The eigenvector at the th principal eigenvector in the th ... The components on each receiving end;

[0062] The dynamic feature descriptor Construct according to the following formula:

[0063]

[0064] The dynamic feature descriptor As part of a fine-grained feature combining the time and frequency domains, it is used to characterize the coupling relationship between the target's motion state and spatial position.

[0065] A passive positioning device based on WIFI wireless signals, used to implement the above method, includes the following steps:

[0066] The signal acquisition module includes: a multi-channel synchronous acquisition unit, used to synchronously acquire WiFi data packets from the same transmitter at multiple receivers and record the original channel state information of each data packet; a complex matrix construction unit, used to extract the amplitude and phase information of each data packet to form an initial complex matrix; a time alignment unit, used to timestamp align the initial complex matrix, generate a time alignment parameter matrix, and construct an initial signal sample set; and a sample filtering unit, used to perform preliminary filtering on the initial signal sample set and remove samples with a signal-to-noise ratio lower than a preset threshold.

[0067] The signal preprocessing module is used to preprocess the filtered signal sample set, remove outliers and noise, and generate a standardized signal sequence.

[0068] The feature extraction module includes: a wavelet transform unit for performing wavelet transform on the standardized signal sequence to extract multi-scale energy distribution features; a phase compensation unit for receiving the time alignment parameter matrix transmitted by the time alignment unit, performing phase compensation on the multi-scale energy distribution features according to the time alignment parameter matrix, and generating a phase-compensated feature matrix; a rotation correction unit for constructing a phase compensation matrix based on the time alignment parameter matrix, performing complex-domain rotation correction on the multi-scale energy distribution features, and generating a rotation-corrected feature matrix containing micro-Doppler phase features; a weighted enhancement unit for constructing a weighted coefficient matrix based on the gradient of the time alignment parameter matrix along the time dimension, performing weighted calculation on the covariance matrix, enhancing the phase components related to the target motion state, and suppressing the phase components corresponding to static multipath interference; a feature decomposition unit for performing feature decomposition on the weighted covariance matrix, extracting principal eigenvalues ​​and their corresponding principal eigenvectors, and constructing a dynamic feature descriptor; and a feature fusion unit for constructing fine-grained features jointly in the time and frequency domains based on the dynamic feature descriptor, and performing normalization processing to form a feature space.

[0069] The localization model module includes: a model storage unit for storing a pre-trained localization model; and a position estimation unit for inputting the feature space into the pre-trained localization model and outputting the position estimation result of the target.

[0070] The feedback optimization module includes: a boundary update unit, used to dynamically update the decision boundary of the positioning model based on the location estimation result; and an iterative control unit, used to feed back the updated decision boundary to the feature extraction module to iteratively optimize feature extraction.

[0071] The output module is used to output the final positioning result.

[0072] A passive positioning device and medium based on WIFI wireless signals, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above method.

[0073] The passive positioning medium based on WIFI wireless signals, when the computer program is executed by the processor, implements the steps of the above method.

[0074] Compared with existing technologies, this invention generates a time alignment parameter matrix by timestamping an initial complex matrix and uses it for phase compensation. This effectively extracts phase offset information introduced by differences in signal propagation paths, overcoming the inconsistency of phase information caused by time asynchrony among multiple receivers in traditional methods. Furthermore, this invention constructs a phase compensation matrix based on the time alignment parameter matrix for complex domain rotation correction, achieving accurate capture of micro-Doppler phase features caused by target motion, enabling the positioning features to reflect the dynamic changes of the target. Moreover, this invention constructs a weighted coefficient matrix based on the gradient of the time alignment parameter matrix to weight the covariance matrix, enhancing the phase components related to target motion and suppressing static multipath interference, significantly highlighting the phase change features caused by target motion. Finally, this invention constructs a dynamic feature descriptor by performing eigenvalue decomposition on the weighted covariance matrix, achieving a joint representation of the target's motion state and spatial position. This allows the positioning model to accurately distinguish targets with different positions and motion states in complex multipath environments, thus significantly improving the accuracy of passive positioning. Attached Figure Description

[0075] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0076] Figure 1 This is a schematic diagram of the overall process provided for an embodiment of the present invention. Detailed Implementation

[0077] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.

[0078] As attached Figure 1 :

[0079] Example 1:

[0080] This invention provides a passive positioning method based on WIFI wireless signals, comprising the following steps:

[0081] S1: Collect WiFi channel status information in the environment and construct an initial signal sample set;

[0082] S2: Preprocess the initial signal sample set to remove outliers and noise, and generate a standardized signal sequence;

[0083] S3: Based on the standardized signal sequence, extract fine-grained features related to the target location and construct a feature space;

[0084] S4: Input the feature space into the pre-trained localization model and output the target's position estimation result;

[0085] S5: Based on the location estimation results, dynamically update the decision boundary of the localization model and feed it back to step S3 to iteratively optimize feature extraction;

[0086] S6: Output the final localization result.

[0087] Step S1 further includes:

[0088] S11: Simultaneously collect WiFi data packets from the same transmitter at multiple receiving ends and record the original channel state information of each data packet;

[0089] S12: Extract the amplitude and phase information of each data packet to form an initial complex matrix;

[0090] S13: Timestamp-align the initial complex matrix to construct the initial signal sample set;

[0091] S14: Perform preliminary screening on the initial signal sample set and remove samples with a signal-to-noise ratio lower than a preset threshold.

[0092] Furthermore, step S3 further includes:

[0093] S31: Perform wavelet transform on the standardized signal sequence to extract multi-scale energy distribution features;

[0094] S32: Principal component analysis is used to reduce the dimensionality of the multi-scale energy distribution features to generate principal component feature vectors;

[0095] S33: Based on the principal component feature vector, construct fine-grained features that combine the time domain and frequency domain;

[0096] S34: Normalize the fine-grained features to form the feature space.

[0097] Specifically, in step S13, when performing timestamp alignment on the initial complex matrix, a time alignment parameter matrix is ​​generated:

[0098]

[0099] in, For the number of receivers, For the number of data packets, This represents the offset relative to the reference timestamp when the nth receiver receives the i-th data packet;

[0100] Wherein, the time alignment parameter matrix Passed to step S32;

[0101] In step S32, during the dimensionality reduction process using principal component analysis, the time alignment parameter matrix is ​​used as a reference. Phase compensation is performed on the multi-scale energy distribution features so that the extracted principal component feature vectors contain phase shift information introduced by the differences in signal propagation paths;

[0102] The phase compensation is performed according to the following formula:

[0103]

[0104] in, These are the complex coefficients corresponding to the multi-scale energy distribution characteristics. The center frequency of the WIFI signal. For the elements in the time alignment parameter matrix, These are the complex coefficients after phase compensation. The natural constant is 2.718.

[0105] It should be noted that the time alignment parameter matrix generated in step S13 The elements in Specifically, it is used to construct the phase compensation matrix in step S32. :

[0106]

[0107] Step S32 is based on the phase compensation matrix. The multi-scale energy distribution features are rotated in the complex domain to ensure that the principal component feature vectors contain micro-Doppler phase features caused by the target motion.

[0108] The elements of the rotated-corrected feature matrix are represented as follows:

[0109]

[0110] in, Represents the complex coefficients after rotation correction. Let be the element in the nth row and ith column of the multi-scale energy distribution feature matrix after wavelet transform.

[0111] Furthermore, the time alignment parameter matrix generated in step S13 The elements in This is further used in step S32 to determine the weighting coefficients of the covariance matrix of the principal component analysis;

[0112] First, calculate the time alignment parameter matrix. Along the gradient of the time dimension, for the nth receiver, the time offset difference between adjacent data packets is:

[0113]

[0114] Constructing the gradient matrix :

[0115]

[0116] Construct a weighted coefficient matrix based on the gradient matrix. :

[0117]

[0118] Among them, diagonal elements Calculate according to the following formula:

[0119]

[0120] in, Preset scale parameters;

[0121] Step S32 is based on the weighting coefficient matrix. Enhance the phase component related to the target motion state and suppress the phase component corresponding to static multipath interference, thereby making the phase change feature caused by the target motion in the principal component feature vector significantly prominent;

[0122] The covariance matrix is ​​calculated using the following weighted formula:

[0123]

[0124] in, The diagonal matrix formed by taking the square roots of each diagonal element of the weighted coefficient matrix is ​​as follows:

[0125]

[0126] Where, (x) T This indicates that the matrix x is transposed.

[0127] Furthermore, the weighted covariance matrix generated in step S32 Passed to step S33;

[0128] In step S33, when constructing the fine-grained features jointly in the time and frequency domains, the weighted covariance matrix is... Perform feature decomposition, extract the principal feature values ​​and their corresponding principal feature vectors, and construct dynamic feature descriptors;

[0129] The eigenvalue decomposition is performed according to the following formula:

[0130]

[0131] in, For the first 1 eigenvalue, Let be the corresponding eigenvectors, and let the eigenvalues ​​be arranged in descending order. ;

[0132] Before selection There are ___ principal eigenvalues ​​and their corresponding principal eigenvectors, where Construct the main feature vector:

[0133]

[0134] Construct the principal feature vector matrix:

[0135]

[0136] in, For the first The eigenvector at the th principal eigenvector in the th ... The components on each receiving end;

[0137] The dynamic feature descriptor Construct according to the following formula:

[0138]

[0139] The dynamic feature descriptor As part of a fine-grained feature combining the time and frequency domains, it is used to characterize the coupling relationship between the target's motion state and spatial position.

[0140] In step S1, WiFi channel state information in the environment is first collected to construct an initial signal sample set. Specifically, WiFi data packets from the same transmitter are simultaneously collected at multiple receivers, and the original channel state information of each data packet is recorded. Subsequently, the amplitude and phase information of each data packet are extracted to form an initial complex matrix. This initial complex matrix is ​​timestamped to construct an initial signal sample set, and the sample set is initially screened to remove samples with a signal-to-noise ratio lower than a preset threshold to ensure the signal quality of subsequent processing.

[0141] In step S2, the initial signal sample set after screening is preprocessed to remove outliers and noise, and a standardized signal sequence is generated, laying the foundation for subsequent feature extraction.

[0142] In step S3, fine-grained features related to the target location are extracted based on the standardized signal sequence to construct a feature space. This step is further refined into several sub-steps.

[0143] First, in step S31, wavelet transform is performed on the standardized signal sequence to extract multi-scale energy distribution features. Wavelet transform can analyze signals simultaneously in the time and frequency domains, effectively capturing the energy variation characteristics of signals at different scales.

[0144] Next, in step S32, principal component analysis is used to reduce the dimensionality of the multi-scale energy distribution features, generating principal component feature vectors. During this process, the time alignment parameter matrix generated in step S13 is implicitly passed to this step. This time alignment parameter matrix records the offset relative to the reference timestamp when each receiver receives each data packet. Phase compensation is performed on the multi-scale energy distribution features based on this time alignment parameter matrix. Specifically, the complex coefficients corresponding to each receiver and each data packet are multiplied by a phase compensation factor, which is determined by the product of the signal center frequency and the corresponding time offset. Through this phase compensation operation, the extracted principal component feature vectors contain phase offset information introduced by differences in signal propagation paths.

[0145] Based on this, a phase compensation matrix is ​​constructed using the time alignment parameter matrix to perform complex-domain rotation correction on the multi-scale energy distribution characteristics. This rotation correction is achieved by multiplying the complex coefficients after wavelet transform by a phase compensation factor, so that the corrected feature matrix includes the micro-Doppler phase characteristics caused by the target motion.

[0146] Furthermore, the gradient of the time alignment parameter matrix along the time dimension is calculated, i.e., the difference in time offset between adjacent data packets at each receiver, to construct a gradient matrix. A weighted coefficient matrix is ​​then constructed based on this gradient matrix, with its diagonal elements obtained by exponentially calculating the ratio of the sum of squares corresponding to each receiver in the gradient matrix to a preset scale parameter. This weighted coefficient matrix is ​​used to weight the covariance matrix, enhancing the phase components related to the target's motion state and suppressing the phase components corresponding to static multipath interference, thereby significantly highlighting the phase change features caused by target motion in the principal component eigenvectors.

[0147] Then, in step S33, a fine-grained feature combining time and frequency domains is constructed based on the generated principal component eigenvectors. Specifically, eigenvalue decomposition is performed on the weighted covariance matrix to extract principal eigenvalues ​​and their corresponding principal eigenvectors. The first few principal eigenvalues ​​and their corresponding principal eigenvectors are selected to construct a principal eigenvalue vector and a principal eigenvector matrix. Subsequently, each principal eigenvalue is multiplied by its corresponding principal eigenvector component to construct a dynamic feature descriptor. This dynamic feature descriptor, as part of the fine-grained feature combining time and frequency domains, is used to characterize the coupling relationship between the target's motion state and spatial position.

[0148] In step S34, the above fine-grained features are normalized to form the final feature space, ensuring that each feature has the same scale, which facilitates subsequent model processing.

[0149] In step S4, the constructed feature space is input into the pre-trained localization model, which outputs the target's position estimation result. This localization model is pre-trained by learning the mapping relationship between the features of a large number of samples and their corresponding positions.

[0150] In step S5, the decision boundary of the localization model is dynamically updated based on the location estimation results, and the updated decision boundary is fed back to step S3 to iteratively optimize the feature extraction process. This feedback mechanism allows for gradual adjustment of the feature extraction direction, making it more adaptable to the current environment and target state.

[0151] Finally, in step S6, the final positioning result is output, completing the entire passive positioning process based on WiFi wireless signals.

[0152] Through the above-described process, this invention integrates time alignment parameters into multiple stages such as phase compensation, rotation correction, weighted enhancement, and feature decomposition, achieving deep fusion of signal propagation path information and target motion information, and effectively improving the accuracy of passive positioning.

[0153] Example 2:

[0154] This embodiment is basically the same as the previous embodiment, except that the passive positioning device based on WIFI wireless signals is used to implement the steps of the above method, including:

[0155] The signal acquisition module includes: a multi-channel synchronous acquisition unit, used to synchronously acquire WiFi data packets from the same transmitter at multiple receivers and record the original channel state information of each data packet; a complex matrix construction unit, used to extract the amplitude and phase information of each data packet to form an initial complex matrix; a time alignment unit, used to timestamp align the initial complex matrix, generate a time alignment parameter matrix, and construct an initial signal sample set; and a sample filtering unit, used to perform preliminary filtering on the initial signal sample set and remove samples with a signal-to-noise ratio lower than a preset threshold.

[0156] The signal preprocessing module is used to preprocess the filtered signal sample set, remove outliers and noise, and generate a standardized signal sequence.

[0157] The feature extraction module includes: a wavelet transform unit for performing wavelet transform on the standardized signal sequence to extract multi-scale energy distribution features; a phase compensation unit for receiving the time alignment parameter matrix transmitted by the time alignment unit, performing phase compensation on the multi-scale energy distribution features according to the time alignment parameter matrix, and generating a phase-compensated feature matrix; a rotation correction unit for constructing a phase compensation matrix based on the time alignment parameter matrix, performing complex-domain rotation correction on the multi-scale energy distribution features, and generating a rotation-corrected feature matrix containing micro-Doppler phase features; a weighted enhancement unit for constructing a weighted coefficient matrix based on the gradient of the time alignment parameter matrix along the time dimension, performing weighted calculation on the covariance matrix, enhancing the phase components related to the target motion state, and suppressing the phase components corresponding to static multipath interference; a feature decomposition unit for performing feature decomposition on the weighted covariance matrix, extracting principal eigenvalues ​​and their corresponding principal eigenvectors, and constructing a dynamic feature descriptor; and a feature fusion unit for constructing fine-grained features jointly in the time and frequency domains based on the dynamic feature descriptor, and performing normalization processing to form a feature space.

[0158] The localization model module includes: a model storage unit for storing a pre-trained localization model; and a position estimation unit for inputting the feature space into the pre-trained localization model and outputting the position estimation result of the target.

[0159] The feedback optimization module includes: a boundary update unit, used to dynamically update the decision boundary of the positioning model based on the location estimation result; and an iterative control unit, used to feed back the updated decision boundary to the feature extraction module to iteratively optimize feature extraction.

[0160] The output module is used to output the final location.

[0161] After the device is started, the signal acquisition module first enters the working state. The multi-channel synchronous acquisition unit synchronously acquires WiFi data packets from the same transmitter at multiple receivers, recording the original channel state information of each data packet. The complex matrix construction unit extracts the amplitude and phase information of each data packet to form an initial complex matrix. The time alignment unit timestamps the initial complex matrix, generates a time alignment parameter matrix, and constructs an initial signal sample set. This time alignment parameter matrix records the offset of each receiver from the reference timestamp when it receives each data packet. The sample filtering unit performs preliminary filtering on the initial signal sample set, removing samples with a signal-to-noise ratio lower than a preset threshold to ensure the signal quality for subsequent processing.

[0162] The signal preprocessing module receives the filtered signal sample set, preprocesses it to remove outliers and noise, generates a standardized signal sequence, and then passes the sequence to the feature extraction module.

[0163] The various units within the feature extraction module work collaboratively according to a pre-defined process. The wavelet transform unit first performs wavelet transform on the standardized signal sequence to extract multi-scale energy distribution features, achieving a preliminary analysis of the signal's time-frequency characteristics.

[0164] The phase compensation unit receives the time alignment parameter matrix implicitly transmitted by the time alignment unit and performs phase compensation on the multi-scale energy distribution characteristics based on this matrix. Specifically, the phase compensation unit multiplies the complex coefficients corresponding to each receiver and each data packet by a phase compensation factor, which is determined by the product of the signal center frequency and the corresponding time offset, to generate a phase-compensated feature matrix. Through this operation, the device ensures that the subsequently extracted features include phase offset information introduced by differences in the signal propagation path.

[0165] The rotation correction unit constructs a phase compensation matrix based on the time alignment parameter matrix, performs complex domain rotation correction on the multi-scale energy distribution characteristics, and multiplies the complex coefficients after wavelet transform by the phase compensation factor to generate a rotation correction feature matrix containing micro-Doppler phase characteristics.

[0166] The weighted enhancement unit calculates the gradient of the time alignment parameter matrix along the time dimension, i.e., the difference in time offset between adjacent data packets at each receiver, and constructs a gradient matrix. Based on this gradient matrix, a weighting coefficient matrix is ​​constructed, whose diagonal elements are obtained by exponentially calculating the ratio of the sum of squares corresponding to each receiver in the gradient matrix to a preset scale parameter. The weighted enhancement unit uses this weighting coefficient matrix to perform weighted calculations on the covariance matrix, enhancing the phase components related to the target's motion state and suppressing the phase components corresponding to static multipath interference.

[0167] The eigenvalue decomposition unit performs eigenvalue decomposition on the weighted covariance matrix, extracts the principal eigenvalues ​​and their corresponding principal eigenvectors, selects the first few principal eigenvalues ​​and their corresponding principal eigenvectors, and multiplies each principal eigenvalue with its corresponding principal eigenvector component to construct a dynamic feature descriptor.

[0168] The feature fusion unit constructs fine-grained features that combine time and frequency domains based on the dynamic feature descriptor, performs normalization processing to form a feature space, and transmits the processing results to the localization model module.

[0169] In the localization model module, the model storage unit stores the pre-trained localization model. The position estimation unit receives the feature space passed from the feature fusion unit, inputs it into the localization model, and outputs the target's position estimation result.

[0170] The feedback optimization module dynamically updates the decision boundary of the positioning model based on the location estimation results. The iterative control unit feeds back the updated decision boundary to the feature extraction module to iteratively optimize the feature extraction process, enabling the device to gradually adapt to environmental changes and target motion states.

[0171] The output module receives the position estimation result from the positioning model module and outputs it as the final positioning result, thus completing a full positioning process.

[0172] During the above process, the time alignment parameter matrix generated by the time alignment unit is implicitly passed to the phase compensation unit, rotation correction unit and weighted enhancement unit, realizing the deep fusion of signal propagation path information and target motion information, enabling the device to achieve accurate positioning in complex dynamic environments.

[0173] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A passive positioning method based on WIFI wireless signals, characterized in that, Includes the following steps: S1: Collect WiFi channel status information in the environment and construct an initial signal sample set; S2: Preprocess the initial signal sample set to remove outliers and noise, and generate a standardized signal sequence; S3: Based on the standardized signal sequence, extract fine-grained features related to the target location and construct a feature space; S4: Input the feature space into the pre-trained localization model and output the target's position estimation result; S5: Based on the location estimation results, dynamically update the decision boundary of the localization model and feed it back to step S3 to iteratively optimize feature extraction; S6: Output the final localization result.

2. The passive positioning method based on WIFI wireless signals according to claim 1, characterized in that, Step S1 further includes: S11: Simultaneously collect WiFi data packets from the same transmitter at multiple receiving ends and record the original channel state information of each data packet; S12: Extract the amplitude and phase information of each data packet to form an initial complex matrix; S13: Timestamp-align the initial complex matrix to construct the initial signal sample set; S14: Perform preliminary screening on the initial signal sample set and remove samples with a signal-to-noise ratio lower than a preset threshold.

3. The passive positioning method based on WIFI wireless signals according to claim 1, characterized in that, Step S3 further includes: S31: Perform wavelet transform on the standardized signal sequence to extract multi-scale energy distribution features; S32: Principal component analysis is used to reduce the dimensionality of the multi-scale energy distribution features to generate principal component feature vectors; S33: Based on the principal component feature vector, construct fine-grained features that combine the time domain and frequency domain; S34: Normalize the fine-grained features to form the feature space.

4. The passive positioning method based on WIFI wireless signals according to claim 2 or 3, characterized in that, In step S13, when performing timestamp alignment on the initial complex matrix, a timestamp alignment parameter matrix is ​​generated: in, For the number of receivers, For the number of data packets, This represents the offset relative to the reference timestamp when the nth receiver receives the i-th data packet; Wherein, the time alignment parameter matrix Passed to step S32; In step S32, during the dimensionality reduction process using principal component analysis, the time alignment parameter matrix is ​​used as a reference. Phase compensation is performed on the multi-scale energy distribution features so that the extracted principal component feature vectors contain phase shift information introduced by the differences in signal propagation paths; The phase compensation is performed according to the following formula: in, These are the complex coefficients corresponding to the multi-scale energy distribution characteristics. The center frequency of the WIFI signal. For the elements in the time alignment parameter matrix, These are the complex coefficients after phase compensation. The natural constant is 2.

718.

5. The passive positioning method based on WIFI wireless signals according to claim 4, characterized in that, The time alignment parameter matrix generated in step S13 The elements in Specifically, it is used to construct the phase compensation matrix in step S32. : Step S32 is based on the phase compensation matrix. The multi-scale energy distribution features are rotated in the complex domain to ensure that the principal component feature vectors contain micro-Doppler phase features caused by the target motion. The elements of the rotated-corrected feature matrix are represented as follows: in, Represents the complex coefficients after rotation correction. Let be the element in the nth row and ith column of the multi-scale energy distribution feature matrix after wavelet transform.

6. The passive positioning method based on WIFI wireless signals according to claim 5, characterized in that, The time alignment parameter matrix generated in step S13 The elements in This is further used in step S32 to determine the weighting coefficients of the covariance matrix of the principal component analysis; First, calculate the time alignment parameter matrix. Along the gradient of the time dimension, for the nth receiver, the time offset difference between adjacent data packets is: Constructing the gradient matrix : Construct a weighted coefficient matrix based on the gradient matrix. : Among them, diagonal elements Calculate according to the following formula: in, Preset scale parameters; Step S32 is based on the weighting coefficient matrix. Enhance the phase component related to the target motion state and suppress the phase component corresponding to static multipath interference, thereby making the phase change feature caused by the target motion in the principal component feature vector significantly prominent; The covariance matrix is ​​calculated using the following weighted formula: in, The diagonal matrix formed by taking the square roots of each diagonal element of the weighted coefficient matrix is ​​as follows: Where, (x) T This indicates that the matrix x is transposed.

7. The passive positioning method based on WIFI wireless signals according to claim 6, characterized in that, The weighted covariance matrix generated in step S32 Passed to step S33; In step S33, when constructing the fine-grained features jointly in the time and frequency domains, the weighted covariance matrix is... Perform feature decomposition, extract the principal feature values ​​and their corresponding principal feature vectors, and construct dynamic feature descriptors; The eigenvalue decomposition is performed according to the following formula: in, For the first 1 eigenvalue, Let be the corresponding eigenvectors, and let the eigenvalues ​​be arranged in descending order. ; Before selection There are ___ principal eigenvalues ​​and their corresponding principal eigenvectors, where Construct the main feature value vector: Construct the principal feature vector matrix: in, For the first The eigenvector at the th principal eigenvector in the th ... The components on each receiving end; The dynamic feature descriptor Construct according to the following formula: The dynamic feature descriptor As part of a fine-grained feature combining the time and frequency domains, it is used to characterize the coupling relationship between the target's motion state and spatial position.

8. A passive positioning device based on WIFI wireless signals, used to implement the steps of the method according to any one of claims 4 to 7, characterized in that, include: The signal acquisition module includes: a multi-channel synchronous acquisition unit, used to synchronously acquire WiFi data packets from the same transmitter at multiple receivers and record the original channel state information of each data packet; a complex matrix construction unit, used to extract the amplitude and phase information of each data packet to form an initial complex matrix; a time alignment unit, used to timestamp align the initial complex matrix, generate a time alignment parameter matrix, and construct an initial signal sample set; and a sample filtering unit, used to perform preliminary filtering on the initial signal sample set and remove samples with a signal-to-noise ratio lower than a preset threshold. The signal preprocessing module is used to preprocess the filtered signal sample set, remove outliers and noise, and generate a standardized signal sequence. The feature extraction module includes: a wavelet transform unit for performing wavelet transform on the standardized signal sequence to extract multi-scale energy distribution features; a phase compensation unit for receiving the time alignment parameter matrix transmitted by the time alignment unit, performing phase compensation on the multi-scale energy distribution features according to the time alignment parameter matrix, and generating a phase-compensated feature matrix; a rotation correction unit for constructing a phase compensation matrix based on the time alignment parameter matrix, performing complex-domain rotation correction on the multi-scale energy distribution features, and generating a rotation-corrected feature matrix containing micro-Doppler phase features; a weighted enhancement unit for constructing a weighted coefficient matrix based on the gradient of the time alignment parameter matrix along the time dimension, performing weighted calculation on the covariance matrix, enhancing the phase components related to the target motion state, and suppressing the phase components corresponding to static multipath interference; a feature decomposition unit for performing feature decomposition on the weighted covariance matrix, extracting principal eigenvalues ​​and their corresponding principal eigenvectors, and constructing a dynamic feature descriptor; and a feature fusion unit for constructing fine-grained features jointly in the time and frequency domains based on the dynamic feature descriptor, and performing normalization processing to form a feature space. The localization model module includes: a model storage unit for storing a pre-trained localization model; and a position estimation unit for inputting the feature space into the pre-trained localization model and outputting the position estimation result of the target. The feedback optimization module includes: a boundary update unit, used to dynamically update the decision boundary of the positioning model based on the location estimation result; and an iterative control unit, used to feed back the updated decision boundary to the feature extraction module to iteratively optimize feature extraction. The output module is used to output the final positioning result.

9. A passive positioning device and medium based on WIFI wireless signals, characterized in that, It includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the method according to any one of claims 1 to 7.

10. A passive positioning medium based on WIFI wireless signals, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images