Wireless positioning method and electronic device

Abstract

The application provides a wireless positioning method and an electronic device. The wireless positioning method specifically comprises the following steps: collecting current channel state information of a user to be positioned based on a channel estimation technology; performing enhanced preprocessing on the current channel state information to obtain a first information enhanced fingerprint feature map or a second information enhanced fingerprint feature map; in response to determining that the first information enhanced fingerprint feature map is obtained, inputting the first information enhanced fingerprint feature map into a pre-constructed positioning model for regression calculation to obtain user position coordinates. The wireless positioning method and the electronic device provided by the application enhance the collected channel state information through preprocessing, and effectively improve the positioning accuracy.

Description

Technical Field

[0001] This application relates to the field of positioning technology, and in particular to a wireless positioning method and electronic device. Background Technology

[0002] With the rapid development and widespread adoption of communication and Internet of Things (IoT) technologies, a large number of portable mobile terminals have entered people's daily lives. Simultaneously, people rely more heavily on their current location in daily life, and using portable mobile terminals to pinpoint their location has become a reality. The demand for location services permeates all aspects of daily life. Traditional positioning technologies are primarily based on ranging, using measurement information from the base station to the user to estimate basic parameters such as angle of arrival and time of arrival. However, this estimation performance degrades significantly in complex multipath environments or non-line-of-sight scenarios. To address this issue, fingerprint-based channel positioning technology emerged and quickly became an important positioning technology. However, existing fingerprint-based channel positioning technologies face difficulties in acquiring channel fingerprint information in complex multipath environments. Furthermore, the acquired channel fingerprints are prone to data distortion due to noise, multipath effects, and channel compression, severely impacting positioning accuracy and limiting the further development and application of fingerprint-based channel positioning technology. Summary of the Invention

[0003] In view of this, the purpose of this application is to propose a wireless positioning method and electronic device to improve positioning accuracy.

[0004] To achieve the above objectives, this application provides a wireless positioning method, comprising:

[0005] The current channel state information of the user to be located is collected based on channel estimation technology;

[0006] The current channel state information is enhanced and preprocessed to obtain a first information-enhanced fingerprint feature map or a second information-enhanced fingerprint feature map;

[0007] In response to obtaining the first information-enhanced fingerprint feature map, the first information-enhanced fingerprint feature map is input into a pre-constructed positioning model for regression calculation to obtain the user's location coordinates.

[0008] Furthermore, the user's current channel state information includes subcarrier domain information, receiving antenna domain information, and transmitting antenna domain information;

[0009] In response to determining that the first information-enhanced fingerprint feature map has been obtained, the preprocessing step includes:

[0010] At the dimensional level of the transmitting antenna domain information, the absolute values ​​of the discrete Fourier transform are performed on the subcarrier domain information and the receiving antenna domain information layer by layer to obtain the power amplitude information corresponding to each layer.

[0011] The power amplitude information corresponding to each layer is concatenated at the dimensional level of the transmitted antenna domain information to obtain a power amplitude data matrix.

[0012] The power amplitude data matrix is ​​reduced in dimensionality using principal component analysis to obtain a two-dimensional data matrix.

[0013] The two-dimensional data matrix is ​​processed into an image to obtain the first information-enhanced fingerprint feature map.

[0014] Furthermore, the user's current channel state information includes subcarrier domain information, receiving antenna domain information, and transmitting antenna domain information;

[0015] In response to determining that the second information-enhanced fingerprint feature map has been obtained, the preprocessing step includes:

[0016] At the dimensional level of the transmitting antenna domain information, the absolute values ​​of the discrete Fourier transform are performed on the subcarrier domain information and the receiving antenna domain information layer by layer to obtain the power amplitude information corresponding to each layer.

[0017] The power amplitude information corresponding to each layer is concatenated at the dimensional level of the transmitted antenna domain information to obtain a power amplitude data matrix.

[0018] The power amplitude data matrix is ​​processed into an image to obtain the second information-enhanced fingerprint feature map.

[0019] Furthermore, the localization model includes a first shallow feature extraction module, a first deep feature extraction module, an image reconstruction module, and a regression output module. The first deep feature extraction module includes multiple residual blocks and convolutional layers set after the multiple residual blocks. Each residual block includes a six-level cascaded sliding window-based Transformer layer and a convolutional layer. The regression output module includes a fully connected layer.

[0020] The step of inputting the first information-enhanced fingerprint feature map into a pre-constructed positioning model for regression calculation to obtain the user's location coordinates includes:

[0021] The shallow features of the first information-enhanced fingerprint feature map are extracted using the first shallow feature extraction module;

[0022] The deep feature extraction module is used to extract deep features from the first information-enhanced fingerprint feature map, specifically including:

[0023] The first information is extracted layer by layer using the six-level cascaded Transformer layers based on sliding windows to enhance the features of the fingerprint feature map;

[0024] The features extracted by the six-level cascaded sliding window-based Transformer layers are fused using the convolutional layers within the residual block;

[0025] The features fused by the convolutional layers within the multiple residual blocks are aggregated using the convolutional layers placed after the multiple residual blocks to obtain the deep features of the first information-enhanced fingerprint feature map.

[0026] The image reconstruction module utilizes the image super-resolution reconstruction algorithm to enhance the shallow and deep features of the fingerprint feature map using the extracted first information to perform image reconstruction, thereby obtaining the reconstructed fingerprint feature map.

[0027] The reconstructed fingerprint feature map is flattened using the regression output module, and then input into the fully connected layer of the regression output module for regression calculation, outputting the user's location coordinates.

[0028] Further, in response to determining that the second information-enhanced fingerprint feature map has been obtained, the following steps are performed:

[0029] The second information-enhanced fingerprint feature map is input into a pre-built classification model for classification to obtain the target classification result;

[0030] Obtain a pre-built fingerprint database, which contains multiple pre-stored fingerprint feature maps, each of which corresponds to at least one preset location coordinate and at least one preset classification label;

[0031] Extract all pre-stored fingerprint feature maps in the fingerprint database whose preset classification labels are consistent with the target classification result, and use them as fingerprint feature maps to be determined.

[0032] Based on the weighted K-nearest neighbor algorithm, the similarity between the second information-enhanced fingerprint feature map and all the undetermined fingerprint feature maps is calculated respectively. The undetermined fingerprint feature maps corresponding to the first K similarity values ​​are sorted in descending order and selected as the target fingerprint feature maps.

[0033] The user's location coordinates are calculated based on the preset location coordinates and similarity corresponding to the target fingerprint feature map.

[0034] Furthermore, the classification model includes a second shallow feature extraction module, a second deep feature extraction module, and a classification output module. The second deep feature extraction module includes multiple residual blocks and convolutional layers set after the multiple residual blocks. Each residual block includes a six-level cascaded sliding window-based Transformer layer and a convolutional layer. The classification output module includes a fully connected layer.

[0035] The step of inputting the second information-enhanced fingerprint feature map into a pre-built classification model for classification to obtain the target classification result includes:

[0036] The shallow features of the second information-enhanced fingerprint feature map are extracted using the second shallow feature extraction module;

[0037] The second deep feature extraction module extracts deep features from the second information-enhanced fingerprint feature map, specifically including:

[0038] The second information is extracted layer by layer using the six-level cascaded Transformer layers based on sliding windows to enhance the features of the fingerprint feature map;

[0039] The features extracted by the six-level cascaded sliding window-based Transformer layers are fused using the convolutional layers within the residual block;

[0040] The features fused by the convolutional layers within the multiple residual blocks are aggregated using the convolutional layers placed after the multiple residual blocks to obtain the deep features of the second information-enhanced fingerprint feature map;

[0041] The shallow and deep features of the extracted second information-enhanced fingerprint feature map are flattened using the classification output module, and then input into the fully connected layer of the classification output module for classification to obtain the target classification result.

[0042] Further, the step of calculating the user's location coordinates based on the preset location coordinates and similarity corresponding to the target fingerprint feature map includes:

[0043] The similarity of all the target fingerprint feature maps is normalized, and the normalization result is used as a weight coefficient. The weight coefficient is then used as the weight of the corresponding preset location coordinates for weighted calculation to obtain the user location coordinates.

[0044] Furthermore, the method for constructing the fingerprint database includes:

[0045] Determine the location area;

[0046] Collect the pre-stored fingerprint feature map corresponding to each preset position coordinate within the positioning area;

[0047] The location area is divided into multiple cells, and each cell is labeled with a preset category label;

[0048] The multiple preset position coordinates are associated with the multiple cells, and each preset position coordinate corresponds to one cell;

[0049] All the pre-stored fingerprint feature maps and their corresponding preset location coordinates and preset classification labels are stored in a database to obtain the fingerprint database;

[0050] Wherein, the preset location coordinates are the actual location coordinates within the positioning area, and the pre-stored fingerprint feature map is the actual fingerprint feature map corresponding to the actual location coordinates.

[0051] Furthermore, the similarity calculation formula between the second information-enhanced fingerprint feature map and the fingerprint feature map to be determined is as follows:

[0052]

[0053] Where S(G0,G1) represents the similarity between the second information-enhanced fingerprint feature map G0 and the undetermined fingerprint feature map G1, L is the error value, T is the transpose, t∈(1,2,3,4,…,N), N is the total number of receiving antennas, x is the error range, and N c This represents the total number of subcarriers.

[0054] Based on the same inventive concept, this application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement a wireless positioning method.

[0055] As can be seen from the above, the wireless positioning method and electronic device provided in this application, by preprocessing the collected current channel state information, enable the first and second information-enhanced fingerprint feature maps to contain rich multipath information features, while stabilizing the sparse multipath features of the current channel state information; inputting the first information-enhanced fingerprint feature map containing rich multipath information features into the positioning model for regression calculation is beneficial for the positioning model to extract features, realize the effective identification of different path information, and effectively recover distorted information and noise information, thereby improving the calculation accuracy of the positioning model and thus effectively improving the positioning accuracy. Attached Figure Description

[0056] To more clearly illustrate the technical solutions in this application or related technologies, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0057] Figure 1 This is a flowchart of a wireless positioning method according to an embodiment of this application;

[0058] Figure 2 This is a schematic diagram of the positioning area grid division according to an embodiment of this application;

[0059] Figure 3 This is a schematic diagram of a wireless positioning device according to an embodiment of this application;

[0060] Figure 4 This is a flowchart of a wireless positioning method for strong multipath information according to an embodiment of this application.

[0061] Figure 5 This is a flowchart of a wireless positioning method for weak multipath information according to an embodiment of this application.

[0062] Figure 6 This is a schematic diagram of an electronic device according to an embodiment of this application. Detailed Implementation

[0063] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with specific embodiments and the accompanying drawings.

[0064] It should be noted that, unless otherwise defined, the technical or scientific terms used in the embodiments of this application should have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms "first," "second," and similar terms used in the embodiments of this application do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed after the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are only used to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0065] With the rapid development and widespread adoption of communication and Internet of Things (IoT) technologies, a vast array of portable mobile devices have entered people's daily lives. Simultaneously, people are increasingly reliant on their current location, making location tracking via portable mobile devices a reality. The demand for location-based services permeates all aspects of daily life. Compared to outdoor positioning, indoor positioning suffers from slower development due to signal blockage caused by numerous walls. Furthermore, wireless signal propagation in complex indoor environments faces challenges such as shadow fading, random signal distortion, and multipath effects, making it difficult for indoor positioning systems to achieve optimal performance.

[0066] Traditional base station positioning technology primarily relies on ranging, using measurement information from the base station to the user to estimate basic parameters such as angle of arrival and time of arrival. However, this estimation performance degrades significantly in complex multipath environments or non-line-of-sight scenarios. To address this issue, fingerprint-based channel positioning technology emerged and quickly became an important positioning technique. Fingerprint-based channel positioning is a method based on the fact that signals in complex environments exhibit different signal strengths or other characteristics at different locations. It replaces geometric measurements and propagation models with fingerprint features, utilizing the different information formed by wireless signals at various points in space to reduce the negative impact of multipath effects on positioning. The fingerprint links the actual location in the environment with the signal characteristic fingerprint, constructing a mapping relationship between signal characteristics and physical spatial location, thereby achieving positioning. In fingerprint-based channel positioning technology, fingerprint information is crucial. There are various types of channel fingerprints currently collected, with Received Signal Strength (RSS, Really Simple Syndication) and Channel State Information (CSI) being the most widely used. RSS channel fingerprints depend on environmental propagation conditions; if environmental interference is significant, the accuracy of RSS acquisition will be compromised. However, since RSS is the superposition of multiple signals, it is greatly affected by multipath propagation. RSS obtained using a network card is often inaccurate, making high-precision positioning difficult. Moreover, its positioning information only contains one intensity dimension, which contains less information. CSI-based positioning has higher accuracy because it can acquire signals from multiple channels, thus collecting more detailed channel state information. Therefore, it is less affected by multipath effects, and its stability and positioning accuracy are also higher.

[0067] Although fingerprint-based CSI positioning has improved accuracy compared to other positioning technologies, the increasing number of users and rising technological demands have led to requirements for higher speeds, lower latency, and greater stability in communication technologies. Existing fingerprint-based channel positioning technologies face challenges in complex multipath environments due to noise, multipath effects, and channel compression, making CSI acquisition difficult. Even when some CSI is acquired, data distortion is common, severely impacting positioning accuracy and limiting the further development and application of fingerprint-based channel positioning technologies.

[0068] Based on the above, this application provides a wireless positioning method, such as... Figure 1 As shown, by preprocessing the collected current channel state information, the resulting first or second information-enhanced fingerprint feature map contains rich multipath information, thereby improving positioning accuracy. Specifically, this includes:

[0069] S11: Collect the current channel state information of the user to be located based on channel estimation technology;

[0070] S12: Perform enhanced preprocessing on the current channel state information to obtain a first information enhanced fingerprint feature map or a second information enhanced fingerprint feature map;

[0071] S13: In response to determining that the first information-enhanced fingerprint feature map has been obtained, the first information-enhanced fingerprint feature map is input into a pre-built positioning model for regression calculation to obtain the user's location coordinates.

[0072] In this process, by enhancing the preprocessing of the collected current channel state information, the resulting first and second enhanced fingerprint feature maps can contain rich multipath information features, while stabilizing the sparse multipath features of the current channel state information. Inputting the first enhanced fingerprint feature map containing rich multipath information features into the positioning model for regression calculation is beneficial for the positioning model to extract features, effectively identify different path information, and effectively recover distorted and noise information, thereby improving the calculation accuracy of the positioning model and thus effectively improving positioning accuracy.

[0073] The following example illustrates the channel modeling process of S11.

[0074] Assume the transmitting end has M antennas and the receiving base station (i.e., the receiver) has N antennas, for a total of W users. There are multiple paths between the transmitting and receiving ends, including both line-of-sight and non-line-of-sight paths. In this case, due to significant signal attenuation and severe signal dispersion on non-line-of-sight paths, information from line-of-sight paths or paths with power exceeding a threshold is typically prioritized for accuracy. In this application, information from P paths with higher power is selected for calculation, and these P paths are considered as the total number of paths between the transmitting and receiving ends.

[0075] The channel impulse response a of the k-th user on the p-th path p,k The expression is: a p,k =α p,k e(θ p,k ), where k∈(1,2,3,4,…,W), p∈(1,2,3,4,…,P), It is the complex decay of the k-th user on the p-th path, e(θ) p,k ) is the arrival angle of the k-th user on the p-th path, and j is the imaginary unit.

[0076] After obtaining a p,k Then, by weighting the responses on the nth subcarrier, the channel frequency response of the kth user on the nth subcarrier can be obtained, i.e. Where n∈(0,2,3,4,…,N) c -1), N c It is the total number of subcarriers, u p,k The sampling interval is h. k,n It contains information about a single subcarrier.

[0077] The two-dimensional matrix of the k-th user across all receiving antennas and all subcarriers is constructed by concatenating the channel frequency responses corresponding to different receiving antennas, expressed as follows: At this time H k This includes subcarrier domain information and receiving antenna domain information, through H k By concatenating the vectors with the transmitted antenna domain information, the current channel state information can be obtained.

[0078] In some embodiments, the user's current channel state information includes subcarrier domain information, receiving antenna domain information, and transmitting antenna domain information.

[0079] In some embodiments, in response to determining that the first information-enhanced fingerprint feature map is obtained, the preprocessing step includes:

[0080] S1211: At the dimensional level of the transmitted antenna domain information, the absolute values ​​of the discrete Fourier transform of the subcarrier domain information and the received antenna domain information are taken layer by layer to obtain the power amplitude information corresponding to each layer.

[0081] S1212: The power amplitude information corresponding to each layer is spliced ​​together at the dimensional level of the transmitted antenna domain information to obtain a power amplitude data matrix;

[0082] S1213: The power amplitude data matrix is ​​reduced in dimensionality using principal component analysis to obtain a two-dimensional data matrix;

[0083] S1214: Perform image processing on the two-dimensional data matrix to obtain the first information-enhanced fingerprint feature map.

[0084] In some embodiments, in response to determining that the second information-enhanced fingerprint feature map has been obtained, the preprocessing step includes:

[0085] S1221: At the dimensional level of the transmitted antenna domain information, the absolute values ​​of the discrete Fourier transform of the subcarrier domain information and the received antenna domain information are taken layer by layer to obtain the power amplitude information corresponding to each layer.

[0086] S1222: The power amplitude information corresponding to each layer is spliced ​​together at the dimension level of the transmitted antenna domain information to obtain a power amplitude data matrix;

[0087] S1223: Perform image processing on the power amplitude data matrix to obtain the second information-enhanced fingerprint feature map.

[0088] Because the strength of multipath information varies significantly under different multipath environments—indoor scenes typically have strong multipath information, while outdoor scenes or scenes with severe overall environmental obstruction usually have weak multipath information—this application employs different preprocessing methods for strong and weak multipath information to improve positioning accuracy. The strength of multipath information can be determined by the scene; for example, outdoor scenes or scenes with severe overall environmental obstruction are considered to have weak multipath information, while indoor scenes or scenes with less overall environmental obstruction are considered to have strong multipath information. Alternatively, it can be determined by the number of reachable paths between a single antenna receiver and transmitter. For instance, a number of reachable paths of 5 or less is considered weak multipath information, while a number of reachable paths greater than 5 is considered strong multipath information. It should be noted that a reachable path count of 5 is not the sole criterion for judging the strength of multipath information; it can be set according to the specific multipath scenario.

[0089] Specifically, when the multipath information is strong, preprocessing steps S1211 to S1214 are performed to obtain the first information-enhanced fingerprint feature map; when the multipath information is weak, preprocessing steps S1221 to S1223 are performed to obtain the second information-enhanced fingerprint feature map.

[0090] Since the obtained current channel state information is a three-dimensional data containing subcarrier domain information, receiving antenna domain information, and transmitting antenna domain information, and the Discrete Fourier Transform (DFT) is difficult to process three-dimensional data simultaneously, in steps S1211 and S1221, the DFT actually processes two-dimensional data composed of subcarrier domain information and receiving antenna domain information. It essentially transforms the processing of three-dimensional data into a multi-level two-dimensional data processing, thus realizing the DFT processing of three-dimensional data. Through the DFT, the subcarrier domain information and receiving antenna domain information are converted into angular domain information and time delay domain information, and then their absolute values ​​are taken to convert them into the corresponding power amplitude information. The specific calculation formula is: Q k =V H H k F, Q k Let V be the composite matrix of the angular domain information and the time delay domain information of the k-th user, where V is the angular domain information, H is the transpose, F is the time delay domain information, and Ω is the time delay domain information. k This represents the power amplitude information for the k-th user. In the Discrete Fourier Transform (DFT), on the one hand, the processed information contains rich multipath information, which is beneficial for extracting shallow and deep features in subsequent localization models; on the other hand, the signal whose absolute value is taken after the DFT changes from two layers of real and imaginary signals to a single layer of real power signal, reducing the number of channels and thus lowering the time complexity. Through the DFT, relatively weaker paths in the current channel state information can be assigned greater weight, facilitating the identification of information from different paths and preserving the multipath information features in the current channel state information to the maximum extent.

[0091] Since the initial current channel state information is a three-dimensional information, in order to preserve the information from different transmitting antennas, it is necessary to use step S1212 or 1222 to splice the power amplitude information of each layer at the dimensional level of the transmitting antenna domain information to obtain a new three-dimensional power amplitude data matrix.

[0092] When multipath information is strong, the power amplitude data matrix is ​​reduced in dimensionality using principal component analysis (PCA) in step S1213 to obtain a two-dimensional data matrix. Essentially, this converts each multi-antenna user into a single-antenna user by fusing multi-layer data, reducing the computational complexity of the positioning model and effectively improving computational efficiency. This achieves both high-efficiency computation and high-precision positioning. However, when multipath information is weak, the collected current channel state information is also weak. Dimensionality reduction based on PCA can easily lose more information, thus affecting subsequent positioning accuracy. Therefore, when multipath information is weak, the power amplitude data matrix is ​​not reduced in dimensionality. Instead, it is directly image-processed to obtain a second information-enhanced fingerprint feature map. This second information-enhanced fingerprint feature map contains richer multipath information features.

[0093] During preprocessing, the sparse multipath features of the current channel state information are stabilized using Discrete Fourier Transform, and more weights are assigned to relatively weaker paths to facilitate the identification of information from different paths. This maximizes the preservation of multipath information features from the current channel state information and improves subsequent positioning accuracy. Simultaneously, for cases with strong multipath information, principal component analysis is used to reduce the dimensionality of the power amplitude data matrix, lowering the computational complexity of the positioning model and effectively improving computational efficiency. This achieves both high-efficiency computation and high-precision positioning.

[0094] In some embodiments, step S1214 includes: normalizing the data in the two-dimensional data matrix, uniformly quantizing the normalized data to a preset gray level, and obtaining the first information-enhanced fingerprint feature map.

[0095] In some embodiments, step S1223 includes: normalizing the data of the power amplitude data matrix, uniformly quantizing the normalized data to a preset gray level, and obtaining the second information-enhanced fingerprint feature map.

[0096] Depending on the actual needs, the preset gray levels of S1214 and S1223 can be set to 256 levels or less. The higher the number of levels, the higher the accuracy, and the higher the accuracy of the subsequent positioning and classification models.

[0097] In some embodiments, the localization model is a Transformer-based neural network model.

[0098] In some embodiments, the localization model includes a first shallow feature extraction module, a first deep feature extraction module, an image reconstruction module, and a regression output module. The first deep feature extraction module includes multiple residual blocks and convolutional layers disposed after the multiple residual blocks. Each residual block includes a six-level cascaded Transformer layer based on a sliding window and a convolutional layer. The regression output module includes a fully connected layer.

[0099] The step of inputting the first information-enhanced fingerprint feature map into a pre-constructed positioning model for regression calculation to obtain the user's location coordinates includes:

[0100] S131: Extract shallow features of the first information-enhanced fingerprint feature map using the first shallow feature extraction module;

[0101] S132: Extracting deep features of the first information-enhanced fingerprint feature map using the first deep feature extraction module, specifically including:

[0102] S1321: The first information is extracted layer by layer using the six-level cascaded Transformer layers based on sliding windows to enhance the features of the fingerprint feature map;

[0103] S1322: The features extracted by the six-level cascaded sliding window-based Transformer layers are fused using the convolutional layers within the residual block;

[0104] S1323: Aggregate the features fused by the convolutional layers within the multiple residual blocks using the convolutional layer set after the multiple residual blocks to obtain the deep features of the first information-enhanced fingerprint feature map;

[0105] S133: Using the image reconstruction module based on the image super-resolution reconstruction algorithm, the shallow and deep features of the fingerprint feature map are enhanced by the extracted first information to perform image reconstruction, and the reconstructed fingerprint feature map is obtained.

[0106] S134: The reconstructed fingerprint feature map is flattened using the regression output module, and then input into the fully connected layer of the regression output module for regression calculation, and the user's location coordinates are output.

[0107] In some embodiments, the first shallow feature extraction module includes a convolutional layer, and shallow features are extracted through the convolutional layer of the first shallow feature extraction module in step S131. Shallow features are low-frequency information, containing more pixel information and more details. Shallow features are more stable and have less distortion than deep features. Applying the extracted shallow features to subsequent image reconstruction can ensure that the reconstructed fingerprint feature map contains more stable and detailed information, resulting in better overall stability. After extracting the shallow features of the first information-enhanced fingerprint feature map, the deep features of the first information-enhanced fingerprint feature map are extracted. Deep features are high-frequency information, with a larger receptive field and stronger semantic information, which can better describe the overall situation of the first information-enhanced fingerprint feature map.

[0108] The first information-enhanced fingerprint feature map is obtained when the multipath information is strong. In steps S131 and S132, the features of the first information-enhanced fingerprint feature map are extracted. Because the first information-enhanced fingerprint feature map has undergone information enhancement processing after preprocessing, it can contain rich multipath information, which is beneficial for feature extraction. In steps S1321 to S1322, the features of different levels of the first information-enhanced fingerprint feature map can be extracted layer by layer using a six-level cascaded sliding window-based Transformer layer (Swin Transformer layer). The convolutional layer in the residual block can bring the inductive bias Swin (sliding window) of the convolution operation into the network and fuse the features extracted by the six-level cascaded sliding window-based Transformer layer. By setting convolutional layers in the residual block, the translation equivalence can be enhanced and the features between different sliding window-based Transformer layers can be effectively aggregated, laying a better foundation for the feature aggregation of each residual block in the subsequent process. At the same time, the six-level cascade can extract rich features from the channels, improving the stability of the localization model. By introducing a sliding window-based Transformer (Swin Transformer) neural network, both deep and shallow features in the first information-enhanced fingerprint feature map can be extracted, effectively improving the localization effect of the localization model and thus increasing the final localization accuracy.

[0109] In step S133, pre-set super-resolution and dimensional parameters are obtained, and image reconstruction is performed based on an image super-resolution reconstruction algorithm. During the image reconstruction process, the information contained in deep and shallow features can be considered simultaneously, greatly improving the image reconstruction effect. The image super-resolution reconstruction algorithm does not require additional hardware; it can improve image resolution and clarity solely through software methods. Currently, most image super-resolution reconstruction algorithms only aim to maximize the peak signal-to-noise ratio of the reconstructed image, resulting in the reconstructed image lacking high-frequency details and having poor visual effects. Furthermore, while image super-resolution reconstruction algorithms using fusion generative adversarial networks can reconstruct images with relatively rich details, they are prone to producing unrealistic pseudo-textures. In this application, when using the image super-resolution reconstruction algorithm for image reconstruction, the information contained in both the shallow and deep features of the first information-enhanced fingerprint feature map is considered simultaneously. This allows the reconstructed fingerprint feature map to retain more details without easily producing pseudo-textures, thereby fully recovering the noisy and distorted information in the first information-enhanced fingerprint feature map. In step S134, the distorted information and noise in the reconstructed fingerprint feature map are effectively recovered, and the multipath information features are preserved to the maximum extent. The regression calculation is then performed using these features, which effectively improves the positioning accuracy.

[0110] In some embodiments, in response to determining that the second information-enhanced fingerprint feature map has been obtained, the following steps are performed:

[0111] S14: Input the second information-enhanced fingerprint feature map into a pre-built classification model for classification to obtain the target classification result;

[0112] S15: Obtain a pre-built fingerprint database, which contains multiple pre-stored fingerprint feature maps, each of which corresponds to at least one preset location coordinate and at least one preset classification label;

[0113] S16: Extract all pre-stored fingerprint feature maps in the fingerprint database whose preset classification labels are consistent with the target classification result, and use them as fingerprint feature maps to be determined;

[0114] S17: Based on the weighted K-nearest neighbor algorithm, calculate the similarity between the second information-enhanced fingerprint feature map and all the undetermined fingerprint feature maps respectively, sort them in descending order of the similarity value, and select the undetermined fingerprint feature maps corresponding to the first K similarity values ​​as the target fingerprint feature maps;

[0115] S18: Calculate the user's location coordinates based on the preset location coordinates and similarity corresponding to the target fingerprint feature map.

[0116] The second information-enhanced fingerprint feature map is obtained when multipath information is weak. When multipath information is weak, steps S14-S18 are continued for localization. The second information-enhanced fingerprint feature map is classified using a classification model, transforming the localization problem into a classification problem. The target classification result yields undetermined fingerprint feature maps, narrowing down the user's location range. Then, the similarity between the second information-enhanced fingerprint feature map and each undetermined fingerprint feature map is calculated using the weighted K-nearest neighbor algorithm (WKNN algorithm). Higher similarity indicates that the preset location coordinates corresponding to that undetermined fingerprint feature map are closer to the user's location coordinates. This is used to calculate the user's location coordinates, effectively solving the problem of poor localization accuracy caused by weak multipath information, thereby effectively improving localization accuracy. Simultaneously, the pre-processed information enhancement of the second information-enhanced fingerprint feature map contains rich multipath information, improving the classification accuracy of the classification model.

[0117] In some embodiments, K in S17 is 5. In addition to 5, it can also be other values ​​greater than or less than 5, such as 4, 3, 2, 1, 6, 7, 8, etc., depending on the actual situation.

[0118] In some embodiments, the classification model includes a second shallow feature extraction module, a second deep feature extraction module, and a classification output module. The second deep feature extraction module includes multiple residual blocks and convolutional layers disposed after the multiple residual blocks. Each residual block includes six cascaded sliding window-based Transformer layers and convolutional layers. The classification output module includes a fully connected layer. S14 includes:

[0119] S141: Extract shallow features of the second information-enhanced fingerprint feature map using the second shallow feature extraction module;

[0120] S142: Extract deep features of the second information-enhanced fingerprint feature map using the second deep feature extraction module, specifically including:

[0121] S1421: The second information is extracted layer by layer using the six-level cascaded sliding window-based Transformer layers to enhance the features of the fingerprint feature map;

[0122] S1422: The features extracted by the six-level cascaded sliding window-based Transformer layers are fused using the convolutional layers within the residual block;

[0123] S1423: Aggregate the features fused by the convolutional layers within the multiple residual blocks using the convolutional layer set after the multiple residual blocks to obtain the deep features of the second information-enhanced fingerprint feature map;

[0124] S143: The shallow and deep features of the extracted second information-enhanced fingerprint feature map are flattened using the classification output module, and then input into the fully connected layer of the classification output module for classification to obtain the target classification result.

[0125] The second shallow feature extraction module and the second deep feature extraction module have the same structure, function, and role as the first shallow feature extraction module and the first deep feature extraction module. The main difference is that the second shallow feature extraction module and the second deep feature extraction module extract the shallow and deep features of the second information-enhanced fingerprint feature map. Therefore, the functions and roles of the second shallow feature extraction module and the second deep feature extraction module will not be elaborated here. By introducing a sliding window-based Transformer (Swin Transformer) neural network, both deep and shallow features in the second information-enhanced fingerprint feature map can be effectively extracted, which effectively improves the classification effect of the classification model, thereby improving the final localization accuracy.

[0126] In some embodiments, S18 includes: normalizing the similarity of all the target fingerprint feature maps, using the normalization result as a weight coefficient, and using the weight coefficient as the weight of the corresponding preset location coordinates to perform weighted calculation to obtain the user location coordinates.

[0127] The higher the similarity of the target fingerprint feature map, the closer its corresponding preset location coordinates are to the user's actual location coordinates. Therefore, it is given a higher weight, so that the final user location coordinates can be more accurate.

[0128] In some embodiments, the method for constructing the fingerprint database includes: determining a positioning area; collecting pre-stored fingerprint feature images corresponding to each preset location coordinate within the positioning area; dividing the positioning area into multiple cells and labeling each cell with a preset category label; associating the multiple preset location coordinates with the multiple cells, with each preset location coordinate corresponding to one cell; and storing all the pre-stored fingerprint feature images and their corresponding preset location coordinates and preset category labels into a database to obtain the fingerprint database. Wherein, the preset location coordinates are the actual location coordinates within the positioning area, and the pre-stored fingerprint feature images are the actual fingerprint feature images corresponding to the actual location coordinates.

[0129] The above method establishes a one-to-one correspondence between each location within the positioning area and the actual fingerprint features of that location. During the actual positioning process, the similarity between the second information-enhanced fingerprint feature map and pre-stored fingerprint feature maps in the fingerprint database is calculated. This is used to estimate the possible location coordinates represented by the second information-enhanced fingerprint feature map. A higher similarity to the pre-stored fingerprint feature map indicates that the user's location coordinates are closer to the preset location coordinates corresponding to that pre-stored fingerprint feature map, thus achieving the prediction of the user's location coordinates.

[0130] In some embodiments, the fingerprint database is updated periodically or irregularly. When changes occur within the location area, the information of the pre-stored fingerprint feature maps can be updated promptly, thereby ensuring the accuracy of the data stored in the database.

[0131] In some embodiments, the similarity calculation formula between the second information-enhanced fingerprint feature map and the fingerprint feature map to be determined is as follows:

[0132]

[0133] Where S(G0,G1) represents the similarity between the second information-enhanced fingerprint feature map G0 and the undetermined fingerprint feature map G1, L is the error value, T is the transpose, t∈(1,2,3,4,…,N), N is the total number of receiving antennas, x is the error range, and N cThis represents the total number of subcarriers.

[0134] In some embodiments, the localization model is a pre-trained model. The pre-training process of the localization model includes: acquiring multiple first training fingerprint feature maps, each of which corresponds to a training location coordinate; dividing the multiple first training fingerprint feature maps into first training data and first verification data according to a preset ratio; inputting the multiple first training fingerprint feature maps into an initial localization model, performing localization training using the first training data, and verifying the training results using the first verification data to obtain a verification result; continuing to perform localization training using the first training data until a preset index is reached; determining the model weight parameter corresponding to the optimal verification result as the final weight parameter, and applying the final weight parameter to the initial localization model to obtain the localization model.

[0135] In the pre-training process of the localization model, each first training fingerprint feature map used for pre-training corresponds to a training location coordinate. The training location coordinate is the actual location coordinate corresponding to the first training fingerprint feature map. During the localization training process, the initial localization model learns the mapping relationship between the first training fingerprint feature map and its corresponding actual location coordinate until the loss function is less than the threshold or the training rounds reach the preset training round threshold, at which point the training ends.

[0136] In some embodiments, the localization training using the first training data includes: extracting shallow and deep features of the first training fingerprint feature map from the first training data; reconstructing the image using the extracted shallow and deep features of the first training fingerprint feature map based on an image super-resolution reconstruction algorithm to obtain a reconstructed training fingerprint feature map; enhancing the channel correlation of the reconstructed training fingerprint feature map based on an attention mechanism to obtain a channel-correlation-enhanced first training fingerprint feature map; flattening the first training fingerprint feature map and the corresponding channel-correlation-enhanced first training fingerprint feature map after superimposing them, and then inputting them into the fully connected layer of the initial localization model for localization training.

[0137] During localization training, the initial training data contains multiple first training fingerprint feature maps, representing information from multiple locations, and the channel correlation between them is poor. Therefore, an attention mechanism is introduced during training to assign weights to these multiple first training fingerprint feature maps, thereby enhancing channel correlation and improving training performance. To prevent the enhanced first training fingerprint feature maps from deviating too much from the original first training fingerprint feature maps, when inputting them into the initial localization model, the enhanced first training fingerprint feature maps are superimposed on their corresponding original first training fingerprint feature maps, flattened, and then input into the fully connected layer of the initial localization model for localization calculation. This training method results in better training performance, better learning performance of the initial localization model, and better performance of the final weight parameters, leading to a higher-performing localization model.

[0138] In some embodiments, the classification model is a pre-trained model. The pre-training process of the classification model includes: acquiring multiple second training fingerprint feature maps, each second training fingerprint feature map corresponding to a training location coordinate and a training classification label, wherein the training classification label is determined based on the training location coordinate; dividing the multiple second training fingerprint feature maps into second training data and second verification data according to a preset ratio; inputting the multiple second training fingerprint feature maps into an initial classification model, performing classification training using the second training data, and verifying the training results using the second verification data to obtain a verification result; continuing to perform classification training using the second training data until a preset index is reached; determining the model weight parameter corresponding to the optimal verification result as the final weight parameter, and applying the final weight parameter to the initial classification model to obtain the classification model.

[0139] In the pre-training of the classification model and the localization model, the preset metrics are a loss function less than a threshold or the number of training epochs reaching a preset training epoch threshold. The preset ratio can be 8:2, meaning 80% of the first or second training fingerprint feature maps are assigned to the first or second training data, and 20% are assigned to the first or second validation data. However, other ratios can be set according to actual circumstances, without specific limitations. During the pre-training process of the classification model, each second training fingerprint feature map used for pre-training corresponds to a training location coordinate and a training classification label. The training location coordinates are the actual location coordinates corresponding to the second training fingerprint feature map. During classification training using the second training data, the initial classification model learns the mapping relationship between the second training fingerprint feature map and its corresponding training classification label until the loss function is less than the threshold or the number of training epochs reaches the preset training epoch threshold, at which point training ends.

[0140] In some embodiments, the classification training using the second training data includes: extracting shallow and deep features from the second training fingerprint feature map in the second training data; enhancing the channel correlation of the shallow and deep features of the second training fingerprint feature map based on an attention mechanism to obtain a channel-correlation-enhanced second training fingerprint feature map; flattening the second training fingerprint feature map and the corresponding channel-correlation-enhanced second training fingerprint feature map by superimposing them, and then inputting them into the fully connected layer of the initial classification model for classification training. The role of the attention mechanism in the classification training process is the same as its role in the pre-training process of the localization model, and will not be elaborated here.

[0141] In some embodiments, training classification labels can be annotated onto the second trained fingerprint feature map using the following methods:

[0142] Determine the area to be located, and divide the entire area into grids, for example... Figure 2 As shown, the positioning area can be divided into 100 uniformly sized square units. The specific number and size of the grid units can be determined according to the actual situation and are not limited here. Each unit unit corresponds to a training classification label, for example, using data from 1 to 100 as their training classification labels. The training position coordinates are the actual coordinates of the corresponding second training fingerprint feature map. By mapping the training position coordinates of each second training fingerprint feature map to the square unit unit, for example, if the training position coordinates fall within the range of 1, it is labeled with the training classification label 1, and so on, all second training fingerprint feature maps are labeled in this way. When the training position coordinates fall on the horizontal line of the grid, the training classification label is assigned according to the upper grid cell; if they fall on the vertical line of the grid, the training classification label is assigned according to the left grid cell, so that each second training fingerprint feature map corresponds to a training classification label. The preset classification labels of fingerprint feature maps pre-stored in the fingerprint database can also be labeled using the above method.

[0143] The wireless positioning method of this application distinguishes between strong and weak multipath information and adopts different preprocessing and positioning methods for each case to achieve better and more accurate positioning. By preprocessing the collected current channel state information to enhance its information, a first information-enhanced fingerprint feature map and a second information-enhanced fingerprint feature map are obtained. This maximizes the preservation of multipath information features and improves the stability of feature information. Especially when the indoor environment becomes more complex, the Discrete Fourier Transform stabilizes the sparse multipath features in the current channel state information and assigns more weights to relatively weaker paths to facilitate the identification of information from different paths and maximize the preservation of multipath information features. At the same time, by introducing a Transformer-based neural network positioning model and classification model, deep and shallow features can be extracted. This significantly improves positioning accuracy while significantly reducing storage space overhead and training time. It can also better recover the original key information from distorted information, enabling the positioning and classification models to have stable performance in complex channel environments. Moreover, the shorter training time and update frequency allow them to better adapt to some volatile environments and are less prone to positioning inaccuracies caused by outdated channel information. This application achieves high-precision wireless positioning, which can be applied in MIMO (Multiple Input Multiple Output) scenarios. It is suitable not only for complex indoor environments but also for complex outdoor environments, with a wide range of applications and high positioning accuracy.

[0144] It should be noted that the method of this application embodiment can be executed by a single device, such as a computer or server. The method of this embodiment can also be applied in a distributed scenario, where multiple devices cooperate to complete the task. In such a distributed scenario, one of these devices may execute only one or more steps of the method of this application embodiment, and the multiple devices will interact with each other to complete the method described. It should be noted that some embodiments of this application have been described above. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in a different order than in the above embodiments and still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some implementations, multitasking and parallel processing are also possible or may be advantageous.

[0145] Based on the same inventive concept, corresponding to any of the above embodiments, this application also provides a wireless positioning device.

[0146] refer to Figure 3 The wireless positioning device includes:

[0147] Information acquisition module 31 is used to acquire the current channel state information of the user to be located based on channel estimation technology;

[0148] Preprocessing module 32 is used to perform enhanced preprocessing on the current channel state information to obtain a first information enhanced fingerprint feature map or a second information enhanced fingerprint feature map;

[0149] The first positioning module 33 is used to, in response to determining that the first information-enhanced fingerprint feature map has been obtained, input the first information-enhanced fingerprint feature map into a pre-constructed positioning model for regression calculation to obtain the user's location coordinates.

[0150] In some embodiments, the user's current channel state information includes subcarrier domain information, receiving antenna domain information, and transmitting antenna domain information; in response to determining that the first information-enhanced fingerprint feature map is obtained, the preprocessing module 32 includes:

[0151] The first discrete Fourier unit is used to perform discrete Fourier transform on the subcarrier domain information and the receiver antenna domain information layer by layer at the dimension level of the transmitted antenna domain information, and then take the absolute value to obtain the power amplitude information corresponding to each layer.

[0152] The first splicing unit is used to splice the power amplitude information corresponding to each layer at the dimension level of the transmitted antenna domain information to obtain a power amplitude data matrix.

[0153] The dimension reduction unit is used to perform dimension reduction processing on the power amplitude data matrix based on principal component analysis to obtain a two-dimensional data matrix.

[0154] The first image processing unit is used to perform image processing on the two-dimensional data matrix to obtain the first information-enhanced fingerprint feature map.

[0155] In some embodiments, the user's current channel state information includes subcarrier domain information, receiving antenna domain information, and transmitting antenna domain information;

[0156] In response to determining that the second information-enhanced fingerprint feature map has been obtained, the preprocessing module 32 includes:

[0157] The second discrete Fourier unit is used to perform discrete Fourier transform on the subcarrier domain information and the receiver antenna domain information layer by layer at the dimension level of the transmitted antenna domain information, and then take the absolute value to obtain the power amplitude information corresponding to each layer.

[0158] The second splicing unit is used to splice the power amplitude information corresponding to each layer at the dimension level of the transmitted antenna domain information to obtain a power amplitude data matrix.

[0159] The second image processing unit is used to perform image processing on the power amplitude data matrix to obtain the second information-enhanced fingerprint feature map.

[0160] In some embodiments, the positioning model includes:

[0161] The first shallow feature extraction module is used to extract shallow features from the first information-enhanced fingerprint feature map;

[0162] The first deep feature extraction module is used to extract deep features from the first information-enhanced fingerprint feature map;

[0163] The image reconstruction module is used to reconstruct the image based on the image super-resolution reconstruction algorithm, using the extracted first information to enhance the shallow and deep features of the fingerprint feature map, and obtain the reconstructed fingerprint feature map.

[0164] The regression output module is used to flatten the reconstructed fingerprint feature map, and then input it into the fully connected layer of the regression output module for regression calculation, and output the user location coordinates.

[0165] In some embodiments, the first deep feature extraction module includes multiple residual blocks and convolutional layers disposed after the multiple residual blocks. Each residual block includes six cascaded sliding window-based Transformer layers and convolutional layers.

[0166] The six-level cascaded Transformer layers based on sliding windows are used to extract features of the first information-enhanced fingerprint feature map layer by layer.

[0167] The convolutional layers within the residual block are used to fuse the features extracted by the six-level cascaded sliding window-based Transformer layers;

[0168] The convolutional layer positioned after the plurality of residual blocks is used to aggregate the features fused by the convolutional layers within the plurality of residual blocks to obtain the deep features of the first information-enhanced fingerprint feature map.

[0169] In some embodiments, a second positioning module is further included, configured to perform the following steps in response to determining that the second information-enhanced fingerprint feature map has been obtained:

[0170] The second information-enhanced fingerprint feature map is input into a pre-built classification model for classification to obtain the target classification result;

[0171] Obtain a pre-built fingerprint database, which contains multiple pre-stored fingerprint feature maps, each of which corresponds to at least one preset location coordinate and at least one preset classification label;

[0172] Extract all pre-stored fingerprint feature maps in the fingerprint database whose preset classification labels are consistent with the target classification result, and use them as fingerprint feature maps to be determined.

[0173] Based on the weighted K-nearest neighbor algorithm, the similarity between the second information-enhanced fingerprint feature map and all the undetermined fingerprint feature maps is calculated respectively. The undetermined fingerprint feature maps corresponding to the first K similarity values ​​are sorted in descending order and selected as the target fingerprint feature maps.

[0174] The user's location coordinates are calculated based on the preset location coordinates and similarity corresponding to the target fingerprint feature map.

[0175] In some embodiments, the classification model includes:

[0176] The second shallow feature extraction module is used to extract shallow features of the second information-enhanced fingerprint feature map.

[0177] The second deep feature extraction module is used to extract deep features of the second information-enhanced fingerprint feature map.

[0178] The classification output module is used to flatten the shallow and deep features of the extracted second information-enhanced fingerprint feature map, and then input them into the fully connected layer of the classification output module for classification to obtain the target classification result.

[0179] In some embodiments, the second deep feature extraction module includes multiple residual blocks and convolutional layers disposed after the multiple residual blocks. Each residual block includes six cascaded sliding window-based Transformer layers and convolutional layers.

[0180] The six-level cascaded Transformer layers based on sliding windows are used to extract the features of the second information-enhanced fingerprint feature map layer by layer;

[0181] The convolutional layers within the residual block are used to fuse the features extracted by the six-level cascaded sliding window-based Transformer layers;

[0182] The convolutional layer placed after the plurality of residual blocks is used to aggregate the features fused by the convolutional layers within the plurality of residual blocks to obtain the deep features of the second information-enhanced fingerprint feature map.

[0183] In some embodiments, the second positioning module is further configured to:

[0184] The similarity of all the target fingerprint feature maps is normalized, and the normalization result is used as a weight coefficient. The weight coefficient is then used as the weight of the corresponding preset location coordinates for weighted calculation to obtain the user location coordinates.

[0185] In some embodiments, a database building module for pre-building a fingerprint database is also included, the database building module comprising:

[0186] Area determination unit, used to determine the range of the positioning area;

[0187] The data acquisition unit is used to acquire the pre-stored fingerprint feature map corresponding to each preset position coordinate within the positioning area;

[0188] The label corresponding unit is used to divide the positioning area into multiple cells, and to label each cell with a preset category label;

[0189] A cell-corresponding unit is used to associate multiple preset position coordinates with multiple cells, with each preset position coordinate corresponding to one cell;

[0190] A storage unit is used to store all the pre-stored fingerprint feature maps and their corresponding preset location coordinates and preset classification labels into a database to obtain the fingerprint database;

[0191] Wherein, the preset location coordinates are the actual location coordinates within the positioning area, and the pre-stored fingerprint feature map is the actual fingerprint feature map corresponding to the actual location coordinates.

[0192] In some embodiments, the similarity calculation formula between the second information-enhanced fingerprint feature map and the fingerprint feature map to be determined is as follows:

[0193]

[0194] Where S(G0,G1) represents the similarity between the second information-enhanced fingerprint feature map G0 and the undetermined fingerprint feature map G1, L is the error value, T is the transpose, t∈(1,2,3,4,…,N), N is the total number of receiving antennas, x is the error range, and N c This represents the total number of subcarriers.

[0195] For ease of description, the above devices are described in terms of function, divided into various modules. Of course, in implementing this application, the functions of each module can be implemented in one or more software and / or hardware.

[0196] The apparatus of the above embodiments is used to implement the corresponding wireless positioning method in any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0197] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, this application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the wireless positioning method described in any of the above embodiments.

[0198] Figure 6 This embodiment illustrates a more specific hardware structure of an electronic device, which may include a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. The processor 1010, memory 1020, input / output interface 1030, and communication interface 1040 are interconnected internally via the bus 1050.

[0199] The processor 1010 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this specification.

[0200] The memory 1020 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 1020 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented by software or firmware, the relevant program code is stored in the memory 1020 and is called and executed by the processor 1010.

[0201] The input / output interface 1030 is used to connect input / output modules to realize information input and output. Input / output modules can be configured as components within the device (not shown in the figure) or externally connected to the device to provide corresponding functions. Input devices may include keyboards, mice, touchscreens, microphones, various sensors, etc., while output devices may include displays, speakers, vibrators, indicator lights, etc.

[0202] The communication interface 1040 is used to connect a communication module (not shown in the figure) to enable communication between this device and other devices. The communication module can communicate via wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0203] Bus 1050 includes a pathway for transmitting information between various components of the device, such as processor 1010, memory 1020, input / output interface 1030, and communication interface 1040.

[0204] It should be noted that although the above-described device only shows the processor 1010, memory 1020, input / output interface 1030, communication interface 1040, and bus 1050, in specific implementations, the device may also include other components necessary for normal operation. Furthermore, those skilled in the art will understand that the above-described device may only include the components necessary for implementing the embodiments of this specification, and not necessarily all the components shown in the figures.

[0205] The electronic devices described above are used to implement the corresponding wireless positioning methods in any of the foregoing embodiments and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0206] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, this application also provides a non-transitory computer-readable storage medium storing computer instructions for causing the computer to execute the wireless positioning method as described in any of the above embodiments.

[0207] The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.

[0208] The computer instructions stored in the storage medium of the above embodiments are used to cause the computer to execute the wireless positioning method as described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0209] It should be noted that the embodiments of this application can be further described in the following ways.

[0210] This application discloses a wireless positioning method, which is applicable when multipath information is strong, such as Figure 4As shown, the following steps are used for positioning:

[0211] S41: Collect the current channel state information of the user to be located based on channel estimation technology.

[0212] S42: Perform enhanced preprocessing on the current channel state information to obtain a first information enhanced fingerprint feature map.

[0213] The user's current channel state information includes subcarrier domain information, receiving antenna domain information, and transmitting antenna domain information; the preprocessing steps include:

[0214] At the dimensional level of the transmitting antenna domain information, the absolute values ​​of the discrete Fourier transform are performed on the subcarrier domain information and the receiving antenna domain information layer by layer to obtain the power amplitude information corresponding to each layer.

[0215] The power amplitude information corresponding to each layer is concatenated at the dimensional level of the transmitted antenna domain information to obtain a power amplitude data matrix.

[0216] The power amplitude data matrix is ​​reduced in dimensionality using principal component analysis to obtain a two-dimensional data matrix.

[0217] The image processing of the two-dimensional data matrix to obtain the first information-enhanced fingerprint feature map specifically includes: normalizing the data in the two-dimensional data matrix, uniformly quantizing the normalized data to a preset gray level, and obtaining the first information-enhanced fingerprint feature map.

[0218] S43: Input the first information-enhanced fingerprint feature map into the pre-built positioning model for regression calculation to obtain the user's location coordinates.

[0219] The localization model includes a first shallow feature extraction module, a first deep feature extraction module, an image reconstruction module, and a regression output module. The first deep feature extraction module includes multiple residual blocks and convolutional layers following the residual blocks. Each residual block includes six cascaded sliding window-based Transformer layers and convolutional layers. The regression output module includes fully connected layers. S43 specifically includes:

[0220] The shallow features of the first information-enhanced fingerprint feature map are extracted using the first shallow feature extraction module;

[0221] The deep feature extraction module is used to extract deep features from the first information-enhanced fingerprint feature map, specifically including:

[0222] The first information is extracted layer by layer using the six-level cascaded Transformer layers based on sliding windows to enhance the features of the fingerprint feature map;

[0223] The features extracted by the six-level cascaded sliding window-based Transformer layers are fused using the convolutional layers within the residual block;

[0224] The features fused by the convolutional layers within the multiple residual blocks are aggregated using the convolutional layers placed after the multiple residual blocks to obtain the deep features of the first information-enhanced fingerprint feature map.

[0225] The image reconstruction module utilizes the image super-resolution reconstruction algorithm to enhance the shallow and deep features of the fingerprint feature map using the extracted first information to perform image reconstruction, thereby obtaining the reconstructed fingerprint feature map.

[0226] The reconstructed fingerprint feature map is flattened using the regression output module, and then input into the fully connected layer of the regression output module for regression calculation, outputting the user's location coordinates.

[0227] This application discloses a wireless positioning method, which is applicable when multipath information is weak, such as Figure 5 As shown, the following steps are used for positioning:

[0228] S51: Collect the current channel state information of the user to be located based on channel estimation technology.

[0229] S52: Perform enhanced preprocessing on the current channel state information to obtain a second information-enhanced fingerprint feature map.

[0230] The user's current channel state information includes subcarrier domain information, receiving antenna domain information, and transmitting antenna domain information; the preprocessing steps include:

[0231] At the dimensional level of the transmitting antenna domain information, the absolute values ​​of the discrete Fourier transform are performed on the subcarrier domain information and the receiving antenna domain information layer by layer to obtain the power amplitude information corresponding to each layer.

[0232] The power amplitude information corresponding to each layer is concatenated at the dimensional level of the transmitted antenna domain information to obtain a power amplitude data matrix.

[0233] The power amplitude data matrix is ​​processed into an image to obtain the second information-enhanced fingerprint feature map. Specifically, this includes: normalizing the data of the power amplitude data matrix, uniformly quantizing the normalized data to a preset gray level, and obtaining the second information-enhanced fingerprint feature map.

[0234] S53: Input the second information-enhanced fingerprint feature map into the pre-built classification model for classification to obtain the target classification result.

[0235] The classification model includes a second shallow feature extraction module, a second deep feature extraction module, and a classification output module. The second deep feature extraction module includes multiple residual blocks and convolutional layers following the residual blocks. Each residual block includes six cascaded sliding window-based Transformer layers and convolutional layers. The classification output module includes fully connected layers. S53 specifically includes:

[0236] The shallow features of the second information-enhanced fingerprint feature map are extracted using the second shallow feature extraction module;

[0237] The second deep feature extraction module extracts deep features from the second information-enhanced fingerprint feature map, specifically including:

[0238] The second information is extracted layer by layer using the six-level cascaded Transformer layers based on sliding windows to enhance the features of the fingerprint feature map;

[0239] The features extracted by the six-level cascaded sliding window-based Transformer layers are fused using the convolutional layers within the residual block;

[0240] The features fused by the convolutional layers within the multiple residual blocks are aggregated using the convolutional layers placed after the multiple residual blocks to obtain the deep features of the second information-enhanced fingerprint feature map;

[0241] The shallow and deep features of the extracted second information-enhanced fingerprint feature map are flattened using the classification output module, and then input into the fully connected layer of the classification output module for classification to obtain the target classification result.

[0242] S54: Obtain a pre-built fingerprint database, which contains multiple pre-stored fingerprint feature maps, each of which corresponds to at least one preset location coordinate and at least one preset classification label.

[0243] The method for constructing the fingerprint database includes: determining the location area; collecting the pre-stored fingerprint feature images corresponding to each preset location coordinate within the location area; dividing the location area into multiple cells and labeling each cell with a preset category label; associating the multiple preset location coordinates with the multiple cells, with each preset location coordinate corresponding to one cell; storing all the pre-stored fingerprint feature images and their corresponding preset location coordinates and preset category labels into a database to obtain the fingerprint database; wherein, the preset location coordinates are the actual location coordinates within the location area, and the pre-stored fingerprint feature images are the actual fingerprint feature images corresponding to the actual location coordinates.

[0244] S55: Extract all pre-stored fingerprint feature maps in the fingerprint database whose preset classification labels are consistent with the target classification result, and use them as fingerprint feature maps to be determined.

[0245] S56: Based on the weighted K-nearest neighbor algorithm, calculate the similarity between the second information-enhanced fingerprint feature map and all the undetermined fingerprint feature maps respectively. Sort the undetermined fingerprint feature maps according to the similarity values ​​from high to low, and select the undetermined fingerprint feature maps corresponding to the top K similarity values ​​as the target fingerprint features. The formula for calculating the similarity between the second information-enhanced fingerprint feature map and the undetermined fingerprint feature maps is as follows:

[0246]

[0247] Where S(G0,G1) represents the similarity between the second information-enhanced fingerprint feature map G0 and the undetermined fingerprint feature map G1, L is the error value, T is the transpose, t∈(1,2,3,4,…,N), N is the total number of receiving antennas, x is the error range, and N c This represents the total number of subcarriers.

[0248] S57: Calculate the user's location coordinates based on the preset location coordinates and similarity corresponding to the target fingerprint feature map.

[0249] The similarity of all the target fingerprint feature maps is normalized, and the normalization result is used as a weight coefficient. The weight coefficient is then used as the weight of the corresponding preset location coordinates for weighted calculation to obtain the user location coordinates.

[0250] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of this application (including the claims) is limited to these examples; within the framework of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of the embodiments of this application as described above, which are not provided in the details for the sake of brevity.

[0251] Additionally, to simplify the description and discussion, and to avoid obscuring the embodiments of this application, the well-known power / ground connections to integrated circuit (IC) chips and other components may or may not be shown in the provided drawings. Furthermore, the apparatus may be shown in block diagram form to avoid obscuring the embodiments of this application, and this also takes into account the fact that the details of the implementation of these block diagram apparatuses are highly dependent on the platform on which the embodiments of this application will be implemented (i.e., these details should be fully understood by those skilled in the art). While specific details (e.g., circuits) have been set forth to describe exemplary embodiments of this application, it will be apparent to those skilled in the art that the embodiments of this application can be implemented without these specific details or with variations thereof. Therefore, these descriptions should be considered illustrative rather than restrictive.

[0252] Although this application has been described in conjunction with specific embodiments thereof, many substitutions, modifications, and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may be used with the embodiments discussed.

[0253] The embodiments of this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the embodiments of this application should be included within the protection scope of this application.

Claims

1. A wireless positioning method, characterized in that, include: The current channel state information of the user to be located is collected based on channel estimation technology; The current channel state information is enhanced and preprocessed to obtain a first information-enhanced fingerprint feature map or a second information-enhanced fingerprint feature map; In response to the current scenario where multipath information is strong, the first information-enhanced fingerprint feature map is obtained; In response to the current scenario where multipath information is weak, the second information-enhanced fingerprint feature map is obtained; In response to determining that the first information-enhanced fingerprint feature map has been obtained, the first information-enhanced fingerprint feature map is input into a pre-constructed positioning model for regression calculation to obtain the user's location coordinates; In response to determining that the second information-enhanced fingerprint feature map has been obtained, the following steps are performed: The second information-enhanced fingerprint feature map is input into a pre-built classification model for classification to obtain the target classification result; Obtain a pre-built fingerprint database, which contains multiple pre-stored fingerprint feature maps, each of which corresponds to at least one preset location coordinate and at least one preset classification label; Extract all pre-stored fingerprint feature maps in the fingerprint database whose preset classification labels are consistent with the target classification result, and use them as fingerprint feature maps to be determined. Based on the weighted K-nearest neighbor algorithm, the similarity between the second information-enhanced fingerprint feature map and all the undetermined fingerprint feature maps is calculated respectively. The undetermined fingerprint feature maps corresponding to the first K similarity values ​​are sorted in descending order and selected as the target fingerprint feature maps. The user's location coordinates are calculated based on the preset location coordinates and similarity corresponding to the target fingerprint feature map; The user's current channel state information includes subcarrier domain information, receiving antenna domain information, and transmitting antenna domain information; In response to determining that the first information-enhanced fingerprint feature map has been obtained, the preprocessing step includes: At the dimensional level of the transmitting antenna domain information, the absolute values ​​of the discrete Fourier transform are performed on the subcarrier domain information and the receiving antenna domain information layer by layer to obtain the power amplitude information corresponding to each layer. The power amplitude information corresponding to each layer is concatenated at the dimensional level of the transmitted antenna domain information to obtain a power amplitude data matrix. The power amplitude data matrix is ​​reduced in dimensionality using principal component analysis to obtain a two-dimensional data matrix. The two-dimensional data matrix is ​​processed into an image to obtain the first information-enhanced fingerprint feature map; In response to determining that the second information-enhanced fingerprint feature map has been obtained, the preprocessing step includes: At the dimensional level of the transmitting antenna domain information, the absolute values ​​of the discrete Fourier transform are performed on the subcarrier domain information and the receiving antenna domain information layer by layer to obtain the power amplitude information corresponding to each layer. The power amplitude information corresponding to each layer is concatenated at the dimensional level of the transmitted antenna domain information to obtain a power amplitude data matrix. The power amplitude data matrix is ​​processed into an image to obtain the second information-enhanced fingerprint feature map.

2. The wireless positioning method according to claim 1, characterized in that, The localization model includes a first shallow feature extraction module, a first deep feature extraction module, an image reconstruction module, and a regression output module. The first deep feature extraction module includes multiple residual blocks and convolutional layers set after the multiple residual blocks. Each residual block includes a six-level cascaded Transformer layer based on a sliding window and a convolutional layer. The regression output module includes a fully connected layer. The step of inputting the first information-enhanced fingerprint feature map into a pre-constructed positioning model for regression calculation to obtain the user's location coordinates includes: The shallow features of the first information-enhanced fingerprint feature map are extracted using the first shallow feature extraction module; The deep feature extraction module is used to extract deep features from the first information-enhanced fingerprint feature map, specifically including: The first information is extracted layer by layer using the six-level cascaded Transformer layers based on sliding windows to enhance the features of the fingerprint feature map; The features extracted by the six-level cascaded sliding window-based Transformer layers are fused using the convolutional layers within the residual block; The features fused by the convolutional layers within the multiple residual blocks are aggregated using the convolutional layers placed after the multiple residual blocks to obtain the deep features of the first information-enhanced fingerprint feature map. The image reconstruction module utilizes the image super-resolution reconstruction algorithm to enhance the shallow and deep features of the fingerprint feature map using the extracted first information to perform image reconstruction, thereby obtaining the reconstructed fingerprint feature map. The reconstructed fingerprint feature map is flattened using the regression output module, and then input into the fully connected layer of the regression output module for regression calculation, outputting the user's location coordinates.

3. The wireless positioning method according to claim 1, characterized in that, The classification model includes a second shallow feature extraction module, a second deep feature extraction module, and a classification output module. The second deep feature extraction module includes multiple residual blocks and convolutional layers set after the multiple residual blocks. Each residual block includes a six-level cascaded Transformer layer based on a sliding window and a convolutional layer. The classification output module includes a fully connected layer. The step of inputting the second information-enhanced fingerprint feature map into a pre-built classification model for classification to obtain the target classification result includes: The shallow features of the second information-enhanced fingerprint feature map are extracted using the second shallow feature extraction module; The second deep feature extraction module extracts deep features from the second information-enhanced fingerprint feature map, specifically including: The second information is extracted layer by layer using the six-level cascaded sliding window-based Transformer layers to enhance the features of the fingerprint feature map; The features extracted by the six-level cascaded sliding window-based Transformer layers are fused using the convolutional layers within the residual block; The features fused by the convolutional layers within the multiple residual blocks are aggregated using the convolutional layers placed after the multiple residual blocks to obtain the deep features of the second information-enhanced fingerprint feature map; The shallow and deep features of the extracted second information-enhanced fingerprint feature map are flattened using the classification output module, and then input into the fully connected layer of the classification output module for classification to obtain the target classification result.

4. The wireless positioning method according to claim 1, characterized in that, The step of calculating the user's location coordinates based on the preset location coordinates and similarity corresponding to the target fingerprint feature map includes: The similarity of all the target fingerprint feature maps is normalized, and the normalization result is used as a weight coefficient. The weight coefficient is then used as the weight of the corresponding preset location coordinates for weighted calculation to obtain the user location coordinates.

5. The wireless positioning method according to claim 1, characterized in that, The method for constructing the fingerprint database includes: Determine the location area; Collect the pre-stored fingerprint feature map corresponding to each preset position coordinate within the positioning area; The location area is divided into multiple cells, and each cell is labeled with a preset category label; The multiple preset position coordinates are associated with the multiple cells, and each preset position coordinate corresponds to one cell; All the pre-stored fingerprint feature maps and their corresponding preset location coordinates and preset classification labels are stored in a database to obtain the fingerprint database; Wherein, the preset location coordinates are the actual location coordinates within the positioning area, and the pre-stored fingerprint feature map is the actual fingerprint feature map corresponding to the actual location coordinates.

6. The wireless positioning method according to claim 1, characterized in that, The similarity calculation formula between the second information-enhanced fingerprint feature map and the fingerprint feature map to be determined is as follows: in, This represents the second information-enhanced fingerprint feature map. With the undetermined fingerprint feature map similarity, L This is the error value. T For transpose, , N Let x be the total number of antennas at the receiving end, and x be the range of error values. This represents the total number of subcarriers.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 6.

Images