Line-of-sight / non-line-of-sight identification assisted positioning method for 6G indoor communication

Abstract

The present invention discloses a line-of-sight / non-line-of-sight identification assisted positioning method suitable for 6G indoor communications, belonging to the field of wireless communications. The method of the present invention inputs the channel fingerprint features of the target positioning point into a pre-built line-of-sight / non-line-of-sight classification model according to the optimal channel fingerprint feature combination rule, outputs the line-of-sight / non-line-of-sight label of the target positioning point, selects a pre-built positioning model and the corresponding optimal channel fingerprint feature combination rule according to the output label type, recombines and arranges the channel fingerprint features of the target positioning point according to the optimal channel fingerprint feature combination rule, inputs the positioning model, and performs adaptive K value weighted prediction to estimate the position coordinates of the target positioning point. The present invention solves the problem of low positioning accuracy caused by a large amount of non-line-of-sight transmission and multipath fading in indoor environments, significantly improves the accuracy and algorithm robustness of indoor positioning, and has potential applicability even in the absence of sufficient training data.

Description

Technical Field

[0001] The present invention relates to the field of wireless communication technologies, and in particular to a line-of-sight / non-line-of-sight identification auxiliary positioning method suitable for 6G indoor communications. Background Art

[0002] The International Telecommunication Union released IMT-2030, designating the 26 GHz, 40 GHz, 60 GHz, and 70 GHz frequency bands as key communication bands for the sixth generation (6G). These are all millimeter wave (mmWave) bands, indicating that mmWave technology holds significant promise for 6G. Compared to sub-6 GHz, mmWave bands offer wider communication bandwidth and higher transmission rates. Furthermore, multiple-input multiple-output (MIMO) technology provides not only delay and power domain information but also additional angular domain information. Due to the shorter wavelength of mmWave, transceiver antennas are often much smaller, making them easier to integrate with MIMO technology. Therefore, the combination of mmWave and MIMO technologies can deliver ultra-high-resolution channel state information (CSI), including angle of arrival, angle of departure, received signal strength (RSS), and time of arrival, potentially meeting the centimeter-level indoor positioning accuracy requirements of 6G.

[0003] Existing indoor positioning methods can be divided into geometric-based positioning methods (GPMs) and fingerprint-based positioning methods (FPMs). GPMs estimate target position using geometric principles based on channel characteristics. However, they are often susceptible to interference from non-line-of-sight (NLoS) transmissions and have strict requirements on transmitter and receiver hardware, resulting in low positioning accuracy and high cost. In contrast, FPMs do not require specialized equipment and can mitigate the interference of NLoS transmissions to a certain extent. FPMs typically involve two phases: offline and online. In the offline phase, FPMs build a channel fingerprint database and train a positioning model. Initial FPMs typically collect RSS as a channel fingerprint feature, but indoor environments often contain more NLoS propagation paths, resulting in severe RSS fluctuations, which affects positioning accuracy and stability. In contrast, CSI provides higher granularity and stability, and can more accurately reflect multipath information. In the online phase, FPMs estimate the actual location of the terminal using the model trained in the offline phase.

[0004] Estimation algorithms used in FPMs can be broadly categorized into three main categories: deterministic, probabilistic, and machine learning. Deterministic algorithms, represented by the weighted K-nearest neighbor (WKNN) algorithm, provide accurate predictions without offline training. However, WKNN fails to reflect the implicit relationship between fingerprints and locations and suffers from poor stability. A representative probabilistic algorithm is the Naive Bayes (NB) estimator, whose positioning accuracy depends heavily on whether the assumed probability distribution matches the true data distribution. In most cases, machine learning algorithms generally outperform the aforementioned two algorithms in terms of prediction accuracy. Furthermore, with the rise of artificial intelligence (AI) in the communications field, an increasing number of researchers are employing artificial neural networks (ANNs) for indoor positioning, such as convolutional neural networks (CNNs), residual networks, and bidirectional long-short-term memory (BiLSTM). However, the accuracy of ANNs relies on a large number of samples, which is challenging for indoor positioning. Furthermore, ANNs are complex in structure, have numerous parameters, and are difficult to train.

[0005] Random forest (RF), an ensemble machine learning algorithm based on decision trees, has demonstrated excellent performance in fields such as AI and classification prediction and has been widely studied. RF can process high-dimensional data without the need for data dimensionality reduction or normalization, and its training speed is fast, making it well-suited for solving multi-classification problems in indoor positioning. However, RF outputs the label with the highest number of votes, making its accuracy highly dependent on a specific label. Furthermore, different training sets exhibit different sensitivities to different features, so using the same feature set in different scenarios can affect the stability of positioning accuracy. Furthermore, while FPMs can partially mitigate the interference of NLoS propagation, they still cannot completely eliminate the impact of NLoS propagation on positioning accuracy. Summary of the Invention

[0006] The present invention provides a line-of-sight / non-line-of-sight identification-assisted positioning method (IAPM) suitable for 6G indoor communications, which solves the problem of low positioning accuracy caused by a large amount of NLoS transmission and multipath fading in indoor environments, and achieves the ultra-high precision and lower latency required by the 6G vision.

[0007] An embodiment of the present invention provides a line-of-sight / non-line-of-sight identification assisted positioning method applicable to 6G indoor communications, comprising the following steps:

[0008] Collecting channel state information of the target positioning point, and extracting a channel fingerprint feature of the target positioning point according to the channel state information;

[0009] Inputting the channel fingerprint features of the target positioning point into a pre-built line-of-sight / non-line-of-sight classification model according to the optimal channel fingerprint feature combination rule, and outputting the line-of-sight / non-line-of-sight label of the target positioning point;

[0010] According to the line-of-sight / non-line-of-sight label of the target positioning point, a pre-built positioning model and the corresponding optimal channel fingerprint feature combination rule are selected. The channel fingerprint features of the target positioning point are recombined and arranged according to the optimal channel fingerprint feature combination rule, input into the positioning model, and adaptive K value weighted prediction is performed to estimate the position coordinates of the target positioning point.

[0011] Optionally, in one embodiment of the present invention, the channel state information of the target positioning point is collected using wireless channel simulation software or wireless channel measurement equipment based on ray tracing, wherein the wireless channel simulation software based on ray tracing collects the channel state information of the target positioning point, including:

[0012] In the ray tracing-based wireless channel simulation software, an indoor scene is constructed and simulation parameters are set. The channel simulation results are output, and each target location can be marked with a line-of-sight or non-line-of-sight label. Each ray in the simulation results corresponds to a set of channel state information, including received signal strength, azimuth arrival angle, elevation arrival angle, and arrival time.

[0013] The channel state information of the target positioning point is collected using a wireless channel measurement device, including: calibrating the wireless channel measurement device, performing channel measurement on the target positioning point and recording the coordinates of the target positioning point and the line-of-sight / non-line-of-sight label; and using a spatial alternating generalized expectation algorithm to perform channel estimation on the channel measurement results to obtain channel state information of each subcarrier.

[0014] Optionally, in one embodiment of the present invention, extracting a channel fingerprint feature of the target positioning point according to the channel state information includes:

[0015] Reconstruct the delay / angular power spectrum density based on the collected channel state information;

[0016] Extracting channel fingerprint features based on the normalized maximum power path, wherein the channel fingerprint features include: received signal strength, azimuth arrival angle, elevation arrival angle and arrival time;

[0017] Channel characteristics are extracted based on the reconstructed delay / angular power spectral density, including received power kurtosis, received power skewness, rise time, RMS delay spread, RMS angular spread, and Ricean K factor.

[0018] Optionally, in one embodiment of the present invention, performing adaptive K-value weighted prediction to estimate the position coordinates of the target positioning point includes:

[0019] After the channel fingerprint features of the target positioning point are rearranged according to the optimal channel fingerprint feature combination and input into the selected positioning model, the scores score_x and score_y of the horizontal and vertical coordinates of the offline reference point in the channel fingerprint database are obtained;

[0020] The K-means clustering algorithm is used to divide the scores score_x and score_y into high-level categories and low-level categories. The number of coordinates contained in the high-level category is the adaptive K value K. x , K y ;

[0021] The K with the highest score x The horizontal axis x k and its score score_x k , K y The vertical coordinate y k and its score score_y k , respectively with the horizontal coordinate score score_x k , vertical coordinate score score_y k is the horizontal coordinate x k , vertical coordinate y k The weight of the target positioning point is obtained by calculating the final prediction coordinates (x T ,y T ):

[0022]

[0023] Where k is K x , K y The index of .

[0024] Optionally, in one embodiment of the present invention, the method further includes: constructing a line-of-sight / non-line-of-sight recognition model, specifically comprising the following steps:

[0025] Establish a reference grid within the positioning area, with the grid lines perpendicular to the coordinate axes of the Cartesian coordinate system. Take the intersection of the grid points as offline reference points, mark the offline reference points with line-of-sight / non-line-of-sight labels, and collect channel status information for the offline reference points.

[0026] Extracting channel fingerprint features of offline reference points based on the channel fingerprint structure, building a channel fingerprint database based on the channel fingerprint features, offline reference point coordinates, and line-of-sight / non-line-of-sight labels, and dividing the channel fingerprint database into a training set and a validation set;

[0027] The channel fingerprint features and line-of-sight / non-line-of-sight labels of the training set are input into the random forest model for training. After executing the feature selection module, the line-of-sight / non-line-of-sight classification model M is output. I And the corresponding optimal channel fingerprint feature combination rule f I ;

[0028] The horizontal coordinate x and the corresponding channel fingerprint feature, the vertical coordinate y and the corresponding channel fingerprint feature of the offline reference point marked as the line-of-sight label in the training set are input into the random forest model for training. After executing the feature selection module, the corresponding positioning model M is output. IP _Lx, M IP _Ly and the corresponding optimal channel fingerprint feature combination rule f IP _Lx, f IP _Ly;

[0029] The horizontal coordinate x and the corresponding channel fingerprint feature, the vertical coordinate y and the corresponding channel fingerprint feature of the offline reference point marked as non-line-of-sight label in the training set are input into the random forest model for training. After executing the feature selection module, the positioning model M is output. IP _Nx, M IP _Ny and the corresponding optimal channel fingerprint feature combination rule f IP _Nx, f IP _Ny.

[0030] Optionally, in one embodiment of the present invention, marking the line-of-sight / non-line-of-sight label of the offline reference point includes:

[0031] If the transmission between the offline reference point and the transmitter is in line of sight, the line of sight / non-line of sight tag of the offline reference point is marked as 1. If the transmission between the offline reference point and the transmitter is in non-line of sight, the line of sight / non-line of sight tag of the offline reference point is marked as 0.

[0032] Optionally, in one embodiment of the present invention, channel state information of an offline reference point is collected using wireless channel simulation software or wireless channel measurement equipment based on ray tracing, wherein collecting channel state information of an offline reference point by the wireless channel simulation software based on ray tracing includes:

[0033] In the ray tracing-based wireless channel simulation software, an indoor scenario is constructed and simulation parameters are set. The channel simulation results are output, along with line-of-sight / non-line-of-sight labels for each offline reference point. Each ray in the simulation results corresponds to a set of channel state information, including received signal strength, azimuth arrival angle, elevation arrival angle, and arrival time.

[0034] Collecting channel state information of offline reference points using wireless channel measurement equipment includes: calibrating the wireless channel measurement equipment, performing channel measurement on each offline reference point and recording the coordinates and line-of-sight / non-line-of-sight labels of the offline reference points; and performing channel estimation on the channel measurement results using a spatial alternating generalized expectation algorithm to obtain channel state information of each subcarrier.

[0035] Optionally, in one embodiment of the present invention, the channel fingerprint features and line-of-sight / non-line-of-sight labels of the training set are input into the random forest model for training. After executing the feature selection module, the line-of-sight / non-line-of-sight classification model M is output. I And the corresponding optimal channel fingerprint feature combination rule f I ,include:

[0036] After the channel fingerprint features and line-of-sight / non-line-of-sight labels of the training set are input into the random forest model for training, the random forest model gives the importance score of the channel fingerprint features;

[0037] According to the importance score of the channel fingerprint feature, sort the channel fingerprint features of the training set in descending order and record the sorting rule f;

[0038] For the channel fingerprint features of the sorted training set, select the first n channel fingerprint features each time and record the sorting rule f of the channel fingerprint features n , input the first n channel fingerprint features selected together with the line-of-sight / non-line-of-sight labels into the random forest model for training and generate n training models, and the channel fingerprint features of the validation set are calculated according to f n After reordering, input the nth training model respectively to estimate the line-of-sight / non-line-of-sight labels of the validation set under the n training models;

[0039] Calculate the consistency ratio of the line-of-sight / non-line-of-sight labels of the n groups of estimated validation sets and the line-of-sight / non-line-of-sight labels of the real validation sets, and get the accuracy of the n training models. Take the training model with the highest accuracy as the line-of-sight / non-line-of-sight classification model M I , line-of-sight / non-line-of-sight classification model M I The corresponding channel fingerprint feature sorting rule is denoted as f I .

[0040] Optionally, in one embodiment of the present invention, the horizontal coordinate x and the corresponding channel fingerprint feature, the vertical coordinate y and the corresponding channel fingerprint feature of the offline reference point marked as the line-of-sight label in the training set are respectively input into the random forest model for training. After executing the feature selection module, the corresponding positioning model M is output. IP _Lx, M IP _Ly and the corresponding optimal channel fingerprint feature combination rule f IP _Lx, f IP _Ly, including:

[0041] After inputting the channel fingerprint features marked as line-of-sight labels in the training set and the corresponding horizontal (vertical) coordinates x(y) into the random forest model for training, the training model gives the channel fingerprint feature importance score;

[0042] According to the importance score of the channel fingerprint feature, sort the channel fingerprint features of the training set in descending order and record the sorting rule f;

[0043] For the sorted training set channel fingerprint features, select the first n channel fingerprint features each time and record the sorting rule f of the channel fingerprint features n , input the first n channel fingerprint features selected together with the corresponding horizontal (vertical) coordinates x(y) into the random forest model for training and generate n training models M x (M y ), the channel fingerprint features marked as line-of-sight labels in the verification set are pressed by f n After reordering, input the nth training model M x (M y ), estimate the horizontal (vertical) coordinate x(y) of the validation set under n training models;

[0044] Combine the estimated horizontal (vertical) coordinates x(y) of the validation set into complete coordinates, and calculate the absolute error between the estimated complete coordinates and the true validation set coordinates. Calculate the average absolute error of n training models, and take the training model with the lowest average absolute error as the positioning model M IP _Lx, M IP _Ly, the channel fingerprint feature sorting rule corresponding to the positioning model is recorded as f IP _Lx, f IP _Ly.

[0045] Optionally, in one embodiment of the present invention, the horizontal coordinate x and the corresponding channel fingerprint feature, the vertical coordinate y and the corresponding channel fingerprint feature of the offline reference point marked as non-line-of-sight label in the training set are respectively input into the random forest model for training. After executing the feature selection module, the positioning model M is output. IP _Nx, M IP _Ny and the corresponding optimal channel fingerprint feature combination rule fIP _Nx, f IP _Ny, including:

[0046] After inputting the channel fingerprint features and the corresponding horizontal (vertical) coordinates x(y) of the training set marked as non-line-of-sight labels into the random forest model for training, the training model gives the channel fingerprint feature importance score;

[0047] According to the importance score of the channel fingerprint feature, sort the channel fingerprint features of the training set in descending order and record the sorting rule f;

[0048] For the sorted training set channel fingerprint features, select the first n channel fingerprint features each time and record the sorting rule f of the channel fingerprint features n , input the first n channel fingerprint features selected together with the corresponding horizontal (vertical) coordinates x(y) into the random forest model for training and generate n training models M x (M y ), the channel fingerprint features marked as non-line-of-sight labels in the verification set are pressed by f n After reordering, input the nth training model M x (M y ), estimate the horizontal (vertical) coordinate x(y) of the validation set under n training models;

[0049] Combine the estimated horizontal coordinate x and vertical coordinate y of the validation set into the complete coordinate, and calculate the absolute error between the estimated complete coordinate and the true validation set coordinate. Then calculate the average absolute error of n training models, and take the training model with the lowest average absolute error as the positioning model M IP _Nx, M IP _Ny, the channel fingerprint feature sorting rule corresponding to the positioning model is recorded as f IP _Nx, f IP _Ny.

[0050] The line-of-sight / non-line-of-sight identification assisted positioning method for 6G indoor communication proposed in an embodiment of the present invention includes a line-of-sight / non-line-of-sight identification module, a feature selection module and an optimized weighted random forest (IWRF) algorithm. Among them, the RF-based IWRF algorithm adopts weighted prediction and adaptive K value to avoid the dependence of the prediction result on a specific label; the feature selection module adaptively adjusts the fingerprint feature sensitivity during the training process to ensure the optimal selection of the fingerprint feature and its accuracy; the line-of-sight / non-line-of-sight identification module adopts the IWRF algorithm to separate line-of-sight and non-line-of-sight data, which greatly reduces the interference of indoor non-line-of-sight transmission on positioning accuracy, while reducing the algorithm complexity. In addition, the present invention also proposes a novel channel fingerprint structure based on channel state information, which includes reconstructing the channel impulse response and extracting channel fingerprint features based on channel state information based on the normalized maximum power (Maximum received power, MRP) path and delay / angular delay power density spectrum. The algorithm proposed in the present invention has high accuracy and robustness, and is superior to existing algorithms. Furthermore, the proposed algorithm demonstrates robustness to environmental variations, demonstrating its potential applicability even without extensive training data. Therefore, the proposed method can meet the centimeter-level indoor positioning accuracy required by 6G systems.

[0051] Additional aspects and advantages of the present invention will be set forth in part in the description which follows and, in part, will be obvious from the description which follows, or may be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0053] Figure 1 Flowchart of a line-of-sight / non-line-of-sight identification assisted positioning method applicable to 6G indoor communications provided in an embodiment of the present invention;

[0054] Figure 2 Schematic diagram of online and offline application scenarios of a line-of-sight / non-line-of-sight identification assisted positioning method applicable to 6G indoor communications according to an embodiment of the present invention;

[0055] Figure 3 A simulation scene graph constructed in Wireless Insite according to an embodiment of the present invention;

[0056] Figure 4 A schematic diagram of a channel measurement scenario according to an embodiment of the present invention;

[0057] Figure 5A two-dimensional distribution diagram of channel measurement scenarios according to an embodiment of the present invention;

[0058] Figure 6 : The spatial distribution diagram of the received power at the transmitting end at (0,0) according to an embodiment of the present invention, (a) is the spatial distribution diagram of the total received power, and (b) is the spatial distribution diagram of the MRP path received power;

[0059] Figure 7 The cumulative error distribution function (CDF) curve of the mean absolute error between the total received power and the MRP path received power using the weighted K-nearest neighbor algorithm according to an embodiment of the present invention is shown;

[0060] Figure 8 A schematic diagram showing a comparison of delay power spectrum density between ray tracing simulation and channel measurement according to an embodiment of the present invention;

[0061] Figure 9 : CDF curves of positioning errors of various algorithms under ray tracing simulation data according to an embodiment of the present invention;

[0062] Figure 10 1 is a CDF curve of positioning errors of various algorithms under actual measurement data according to an embodiment of the present invention. DETAILED DESCRIPTION

[0063] The following describes embodiments of the present invention in detail, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and are not to be construed as limiting the present invention.

[0064] Figure 1 The present invention provides a flowchart of a line-of-sight / non-line-of-sight identification auxiliary positioning method suitable for 6G indoor communications according to an embodiment of the present invention.

[0065] like Figure 1 As shown, the line-of-sight / non-line-of-sight identification assisted positioning method applicable to 6G indoor communication includes the following steps:

[0066] Step S101: collecting channel state information of a target positioning point, and extracting a channel fingerprint feature of the target positioning point according to the channel state information.

[0067] In an embodiment of the present invention, the channel state information of the target positioning point may be collected using wireless channel simulation software or wireless channel measurement equipment based on ray tracing.

[0068] Specifically, the ray tracing-based wireless channel simulation software collects channel state information of the target positioning point, including:

[0069] In the ray tracing-based wireless channel simulation software, an indoor scene is constructed and simulation parameters such as the transmitter position, target positioning position, electromagnetic wave frequency, bandwidth, transmission power, and propagation mechanism are set. In the ray tracing-based wireless channel simulation software, the channel simulation results are output and each target positioning point can be marked with a line-of-sight / non-line-of-sight label. Each ray in the simulation result corresponds to a set of channel state information, including: received signal strength, azimuth arrival angle, pitch arrival angle, and arrival time.

[0070] The channel state information of the target positioning point is collected using a wireless channel measurement device, including: performing channel measurement on the target positioning point after calibrating the wireless channel measurement device and recording the target positioning point coordinates and line-of-sight / non-line-of-sight tags; using the spatial alternating generalized expectation algorithm to perform channel estimation on the channel measurement results to obtain the channel state information of each subcarrier, including: received signal strength, azimuth arrival angle, pitch arrival angle and arrival time.

[0071] Specifically, the channel fingerprint features of the target positioning point are extracted according to the channel state information, including:

[0072] Reconstruct the delay / angular power spectrum density based on the collected channel state information;

[0073] Channel fingerprint features are extracted based on the normalized maximum power path. The channel fingerprint features include: received signal strength, azimuth arrival angle, elevation arrival angle and arrival time;

[0074] Channel characteristics are extracted based on the reconstructed delay / angular power spectral density, including received power kurtosis, received power skewness, rise time, RMS delay spread, RMS angular spread, and Ricean K factor.

[0075] Step S102: input the channel fingerprint features of the target positioning point into a pre-built line-of-sight / non-line-of-sight classification model according to the optimal channel fingerprint feature combination rule, and output the line-of-sight / non-line-of-sight label of the target positioning point.

[0076] Step S103: Based on the line-of-sight / non-line-of-sight label of the target positioning point, a pre-built positioning model and the corresponding optimal channel fingerprint feature combination rule are selected. The channel fingerprint features of the target positioning point are rearranged according to the optimal channel fingerprint feature combination rule, input into the positioning model, and adaptive K value weighted prediction is performed to estimate the position coordinates of the target positioning point.

[0077] Optionally, in one embodiment of the present invention, performing adaptive K-value weighted prediction to estimate the position coordinates of the target positioning point includes:

[0078] After the channel fingerprint features of the target positioning point are rearranged according to the optimal channel fingerprint feature combination and input into the selected positioning model, the scores score_x and score_y of the horizontal and vertical coordinates of the offline reference point in the channel fingerprint database are obtained;

[0079] The K-means clustering algorithm is used to divide the scores score_x and score_y into high-level categories and low-level categories. The number of coordinates contained in the high-level category is the adaptive K value K. x , K y ;

[0080] The K with the highest score x The horizontal axis x k and its score score_x k , K y The vertical coordinate y k and its score score_y k , respectively with the horizontal coordinate score score_x k , vertical coordinate score score_y k is the horizontal coordinate x k , vertical coordinate y k The weight of the target positioning point is obtained by calculating the final prediction coordinates (x T ,y T ):

[0081]

[0082] Where k is K x , K y The index of .

[0083] In an embodiment of the present invention, before collecting the channel state information of the target positioning point, a line-of-sight / non-line-of-sight recognition model is constructed, which specifically includes the following steps:

[0084] Establish a reference grid within the positioning area, with the grid lines perpendicular to the coordinate axes of the Cartesian coordinate system. Take the intersection of the grid points as offline reference points, mark the offline reference points with line-of-sight / non-line-of-sight labels, and collect channel status information for the offline reference points.

[0085] The channel fingerprint features of the offline reference points are extracted based on the channel fingerprint structure, and a channel fingerprint database is constructed based on the channel fingerprint features, the offline reference point coordinates, and the line-of-sight / non-line-of-sight labels. The channel fingerprint database is then divided into a training set and a validation set.

[0086] The channel fingerprint features and line-of-sight / non-line-of-sight labels of the training set are input into the random forest model for training. After executing the feature selection module, the line-of-sight / non-line-of-sight classification model M is output. I And the corresponding optimal channel fingerprint feature combination rule f I;

[0087] The horizontal coordinate x and the corresponding channel fingerprint feature, the vertical coordinate y and the corresponding channel fingerprint feature of the offline reference point marked as the line-of-sight label in the training set are input into the random forest model for training. After executing the feature selection module, the corresponding positioning model M is output. IP _Lx, M IP _Ly and the corresponding optimal channel fingerprint feature combination rule f IP _Lx, f IP _Ly;

[0088] The horizontal coordinate x and the corresponding channel fingerprint feature, the vertical coordinate y and the corresponding channel fingerprint feature of the offline reference point marked as non-line-of-sight label in the training set are input into the random forest model for training. After executing the feature selection module, the positioning model M is output. IP _Nx, M IP _Ny and the corresponding optimal channel fingerprint feature combination rule f IP _Nx, f IP _Ny.

[0089] Optionally, in one embodiment of the present invention, marking the line-of-sight / non-line-of-sight label of the offline reference point includes:

[0090] If the transmission between the offline reference point and the transmitter is in line of sight, the line of sight / non-line of sight tag of the offline reference point is marked as 1. If the transmission between the offline reference point and the transmitter is in non-line of sight, the line of sight / non-line of sight tag of the offline reference point is marked as 0.

[0091] Optionally, in one embodiment of the present invention, channel state information of an offline reference point is collected using wireless channel simulation software or wireless channel measurement equipment based on ray tracing, wherein collecting channel state information of an offline reference point by the wireless channel simulation software based on ray tracing includes:

[0092] In the ray tracing-based wireless channel simulation software, an indoor scenario is constructed and simulation parameters are set. The channel simulation results are output, along with line-of-sight / non-line-of-sight labels for each offline reference point. Each ray in the simulation results corresponds to a set of channel state information, including received signal strength, azimuth arrival angle, elevation arrival angle, and arrival time.

[0093] Collecting channel state information of offline reference points using wireless channel measurement equipment includes: calibrating the wireless channel measurement equipment, performing channel measurement on each offline reference point and recording the coordinates and line-of-sight / non-line-of-sight labels of the offline reference points; and performing channel estimation on the channel measurement results using a spatial alternating generalized expectation algorithm to obtain channel state information of each subcarrier.

[0094] Optionally, in one embodiment of the present invention, the channel fingerprint features and line-of-sight / non-line-of-sight labels of the training set are input into the random forest model for training. After executing the feature selection module, the line-of-sight / non-line-of-sight classification model M is output. I And the corresponding optimal channel fingerprint feature combination rule f I ,include:

[0095] After the channel fingerprint features and line-of-sight / non-line-of-sight labels of the training set are input into the random forest model for training, the random forest model gives the importance score of the channel fingerprint features;

[0096] According to the importance score of the channel fingerprint feature, sort the channel fingerprint features of the training set in descending order and record the sorting rule f;

[0097] For the channel fingerprint features of the sorted training set, select the first n channel fingerprint features each time and record the sorting rule f of the channel fingerprint features n , input the first n channel fingerprint features selected together with the line-of-sight / non-line-of-sight labels into the random forest model for training and generate n training models, and the channel fingerprint features of the validation set are calculated according to f n After reordering, input the nth training model respectively to estimate the line-of-sight / non-line-of-sight labels of the validation set under the n training models;

[0098] Calculate the consistency ratio of the line-of-sight / non-line-of-sight labels of the n groups of estimated validation sets and the line-of-sight / non-line-of-sight labels of the real validation sets, and get the accuracy of the n training models. Take the training model with the highest accuracy as the line-of-sight / non-line-of-sight classification model M I , line-of-sight / non-line-of-sight classification model M I The corresponding channel fingerprint feature sorting rule is denoted as f I .

[0099] Optionally, in one embodiment of the present invention, the horizontal coordinate x and the corresponding channel fingerprint feature, the vertical coordinate y and the corresponding channel fingerprint feature of the offline reference point marked as the line-of-sight label in the training set are respectively input into the random forest model for training. After executing the feature selection module, the corresponding positioning model M is output. IP _Lx, M IP _Ly and the corresponding optimal channel fingerprint feature combination rule f IP _Lx, f IP _Ly, including:

[0100] After inputting the channel fingerprint features marked as line-of-sight labels in the training set and the corresponding horizontal (vertical) coordinates x(y) into the random forest model for training, the training model gives the channel fingerprint feature importance score;

[0101] According to the importance score of the channel fingerprint feature, sort the channel fingerprint features of the training set in descending order and record the sorting rule f;

[0102] For the sorted training set channel fingerprint features, select the first n channel fingerprint features each time and record the sorting rule f of the channel fingerprint features n , input the first n channel fingerprint features selected together with the corresponding horizontal (vertical) coordinates x(y) into the random forest model for training and generate n training models M x (M y ), the channel fingerprint features marked as line-of-sight labels in the verification set are pressed by f n After reordering, input the nth training model M x (M y ), estimate the horizontal (vertical) coordinate x(y) of the validation set under n training models;

[0103] Combine the estimated horizontal (vertical) coordinates x(y) of the validation set into complete coordinates, and calculate the absolute error between the estimated complete coordinates and the true validation set coordinates. Calculate the average absolute error of n training models, and take the training model with the lowest average absolute error as the positioning model M IP _Lx, M IP _Ly, the channel fingerprint feature sorting rule corresponding to the positioning model is recorded as f IP _Lx, f IP _Ly.

[0104] Optionally, in one embodiment of the present invention, the horizontal coordinate x and the corresponding channel fingerprint feature, the vertical coordinate y and the corresponding channel fingerprint feature of the offline reference point marked as non-line-of-sight label in the training set are respectively input into the random forest model for training. After executing the feature selection module, the positioning model M is output. IP _Nx, M IP _Ny and the corresponding optimal channel fingerprint feature combination rule f IP _Nx, f IP _Ny, including:

[0105] After inputting the channel fingerprint features and the corresponding horizontal (vertical) coordinates x(y) of the training set marked as non-line-of-sight labels into the random forest model for training, the training model gives the channel fingerprint feature importance score;

[0106] According to the importance score of the channel fingerprint feature, sort the channel fingerprint features of the training set in descending order and record the sorting rule f;

[0107] For the sorted training set channel fingerprint features, select the first n channel fingerprint features each time and record the sorting rule f of the channel fingerprint features n, input the first n channel fingerprint features selected together with the corresponding horizontal (vertical) coordinates x(y) into the random forest model for training and generate n training models M x (M y ), the channel fingerprint features marked as non-line-of-sight labels in the verification set are pressed by f n After reordering, input the nth training model M x (M y ), estimate the horizontal (vertical) coordinate x(y) of the validation set under n training models;

[0108] Combine the estimated horizontal coordinate x and vertical coordinate y of the validation set into the complete coordinate, and calculate the absolute error between the estimated complete coordinate and the true validation set coordinate. Then calculate the average absolute error of n training models, and take the training model with the lowest average absolute error as the positioning model M IP _Nx, M IP _Ny, the channel fingerprint feature sorting rule corresponding to the positioning model is recorded as f IP _Nx, f IP _Ny.

[0109] Combine Figure 2 As shown, the line-of-sight / non-line-of-sight identification auxiliary positioning method applicable to 6G indoor communication of the present invention is described in detail through a specific embodiment.

[0110] In a specific embodiment, a line-of-sight / non-line-of-sight identification assisted positioning method applicable to 6G indoor communication specifically includes the following steps:

[0111] Step S1, establish a reference grid in the positioning area at intervals of 0.5m×0.5m, with the grid lines perpendicular to the coordinate axes of the Cartesian coordinate system, take the grid intersection as the offline reference point, mark the line-of-sight / non-line-of-sight label of the offline reference point, and collect the channel state information of the offline reference point. In this embodiment, the area of ​​the positioning area is 46m×17m, and there are 12 pillars in the environment. There is a large amount of non-line-of-sight transmission, but line-of-sight transmission is still dominant. In addition, the height of the offline reference point is set to 1.5m, and the height of the transmitter is set to 2.8m. Both the transmitter and the receiver use omnidirectional antennas, the transmitter is a multi-input multi-output array antenna, and the receiver is a single antenna. In this example, the channel state information of the offline reference point is collected by using both the commercial ray tracing simulation software Wireless InSite and the wireless channel measurement equipment. The following sub-steps are specifically included in Wireless InSite:

[0112] Step S101: Set the height of the offline reference point to 1.5m, the height of the transmitter to 2.8m, both the transmitter and the receiver use omnidirectional antennas, the transmitter is a 4×4 array antenna, the receiver is a 2×2 array antenna, the transmitter position coordinates are (13.59m, 14.11m), the offline reference point is located at the intersection of the 0.5m×0.5m grid, the electromagnetic wave frequency is 60GHz, the bandwidth is 3GHz, the transmission power is 23dBm, the maximum allowable reflection order is 6th order, the diffraction order is 1st order, and the constructed scene model is as follows: Figure 3 shown.

[0113] Step S102: In Wireless InSite, output the channel simulation results and each reference point can be labeled with a line-of-sight or non-line-of-sight label. Each ray in the simulation results corresponds to a set of channel state information, including: received signal strength, azimuth arrival angle, elevation arrival angle, and arrival time.

[0114] Collecting channel state information of an offline reference point using a wireless channel measurement device includes the following sub-steps:

[0115] Step S103: Indoor channel measurement scenario Figure 4 As shown, the scene is Figure 3 The Wireless InSite modeling scenario shown in the figure is consistent. The transmitting and receiving antennas are set to 4×8 dual-polarization cylindrical MIMO and 4×4 dual-polarization planar MIMO, with a bandwidth of 700 MHz. After calibrating the wireless channel measurement equipment, channel measurements are performed on each offline reference point, and the offline reference point coordinates and line-of-sight / non-line-of-sight labels are recorded. Channel state information is collected for a total of 130 offline reference points, of which 90 are in the aisle, representing a line-of-sight environment, and 30 are on the table, representing a non-line-of-sight environment. Figure 5 As shown; 80% of the points in the allocated reference points are used to construct the offline database.

[0116] Step S104: perform channel estimation on the channel measurement results using the spatial alternating generalized expectation algorithm to obtain channel state information of each subcarrier, including received signal strength, azimuth arrival angle, elevation arrival angle, and arrival time.

[0117] Step S2: Extract channel fingerprint features based on the novel channel fingerprint structure, and build a channel fingerprint database based on the channel fingerprint features, offline reference point coordinates, and line-of-sight / non-line-of-sight labels. Randomly select part of the channel fingerprint database as a training set and part of the data as a validation set. The novel channel fingerprint structure includes the following sub-steps:

[0118] Step S201: Reconstruct the delay / angle power spectrum density based on the channel state information in step S102 or step S104; after obtaining multipath parameters such as angle, delay, phase, and power, reconstruct the average delay power spectrum of multiple antenna pairs as the delay power spectrum of the MIMO system using the following formula to reduce noise interference:

[0119]

[0120] Among them, Nsc is a constant, which represents the number of subcarriers. Here, it is 1024. T 、N R are the number of array elements of the transmitting and receiving antenna arrays, h pq (τ) is H pq (f) is the inverse Fourier transform, H pq (f) is the channel transfer function between the transmitting antenna element p and the receiving antenna element q, and the calculation formula is:

[0121]

[0122] Where L is the number of multipaths, α l is the multipath amplitude, is the phase, τ l is the multipath delay, and f is the frequency within the bandwidth.

[0123] The angular power spectral density can be obtained by the angles and received powers of the L multipaths:

[0124]

[0125] Where pl represents the power of the lth path, δ(·) represents the Dirac-delta function, and θ is the azimuth arrival angle Φ AAoA 、Azimuth departure angle Φ AAoD , pitch arrival angle θ EAoA and the pitch departure angle θ EAoD ,θ l Φ represents the lth path AAoA , Φ AAoD ,θ EAoA or θ EAoD .

[0126] Step S202: Extract channel fingerprint features based on the MRP path, including: received signal strength, azimuth arrival angle, elevation arrival angle, and arrival time; the MRP path represents the propagation path with the highest normalized power during signal transmission, and contributes the most to the power received at the receiving end. In 6G indoor millimeter wave communication scenarios, the MRP paths at most receiving ends are usually line-of-sight paths. Even if the MRP paths are non-line-of-sight paths, their power loss is still acceptable due to the characteristics of millimeter waves, and they usually only undergo low-order reflections or diffractions. Therefore, the received power intensity of the MRP path can greatly reduce the interference caused by non-line-of-sight propagation, such as Figure 6 In addition, when the deterministic weighted K nearest neighbor algorithm is used for positioning, it can be clearly seen from the cumulative distribution function of the mean absolute error of positioning that the received power intensity of the MRP path significantly improves the relative positioning accuracy, as shown in Figure 7 In addition, from Figure 8 It can be seen that the delay power spectrum density reconstructed from the ray tracing simulation data is basically consistent with the delay power spectrum density of the channel measurement data at the MRP path.

[0127] Step S203: Channel characteristics are extracted based on the reconstructed delay / angular power spectral density (DPAD), including received power kurtosis, received power skewness, rise time, RMS delay spread, RMS angular spread, and Rice's K factor. Extracting only the channel characteristics of the MRP path can lead to the loss of valuable information, especially in non-line-of-sight (NLOS) environments. To address this issue, a channel fingerprint based on the DPAD / angular power spectral density (DPAD) is introduced to compensate for the information loss caused by the MRP path-based channel fingerprint. Furthermore, these multipath characteristics can better enable line-of-sight (LOS) and non-line-of-sight (NLOS) identification.

[0128] 1) Received power kurtosis:

[0129]

[0130] Where E(·) represents the expectation, μ h and σ h yes The mean and standard deviation of can be expressed as:

[0131]

[0132] 2) Received power skewness

[0133]

[0134] 3) Rise time

[0135]

[0136] 4) Root mean square delay spread

[0137]

[0138] Among them, μ τ It can be calculated by the following formula:

[0139]

[0140] 5) RMS angle expansion

[0141]

[0142] in, It can be φ AAoA ,φ AAoD ,θ EAoA or θ EAoD , μ θ It can be calculated by the following formula:

[0143]

[0144] 6) Rice K factor

[0145]

[0146] Among them, G a , G v Can be calculated by the following formula:

[0147]

[0148] Therefore, in the embodiment of the present invention, the offline channel fingerprint database is constructed with a dimension of N RP ×18 matrix, where N RP The first 15 columns are channel fingerprint features, columns 16 and 17 are the geographic coordinates of the offline reference points, and the last column is a binary label representing line-of-sight / non-line-of-sight, where the line-of-sight label is assigned to 1 and the non-line-of-sight label is assigned to 0. Channel fingerprint features include Δτ,σ DS ,σ AS , in, σ DS Including Φ AAoA , Φ AAoD ,θ EAoA , and θ EAoD .

[0149] Step S3: Input the channel fingerprint features and line-of-sight / non-line-of-sight labels of the training set into RF for training. After executing the feature selection module, the line-of-sight / non-line-of-sight classification model M is output. I And the corresponding optimal channel fingerprint feature combination rule f I; The feature selection module includes the following sub-steps:

[0150] Step S301: After the channel fingerprint features and line-of-sight / non-line-of-sight labels of the training set are input into the RF for training, the training model gives the channel fingerprint feature importance score;

[0151] Step S302: sort the channel fingerprint features of the training set in descending order according to the importance scores of the channel fingerprint features, and record the sorting rule f;

[0152] Step S303: For the sorted training set channel fingerprint features, select the first n channel fingerprint features each time and record the sorting rule f of the channel fingerprint features. n , input the first n channel fingerprint features selected together with the line-of-sight / non-line-of-sight labels into RF for training and generate n training models, and the channel fingerprint features of the verification set are calculated according to f n After reordering, input the nth training model respectively to estimate the line-of-sight / non-line-of-sight labels of the validation set under the n training models;

[0153] Step S304: Calculate the consistency ratio of the n estimated validation set line-of-sight / non-line-of-sight labels and the real validation set line-of-sight / non-line-of-sight labels, that is, calculate the accuracy of the n training models, and take the training model with the highest accuracy as M. I , the channel fingerprint feature sorting rule corresponding to this model is recorded as f I .

[0154] Step S4: The horizontal coordinate x and the corresponding channel fingerprint feature, the vertical coordinate y and the corresponding channel fingerprint feature of the offline reference point marked with the line-of-sight label in the training set are input into RF for training. After executing the feature selection module, the positioning model M is output. IP _Lx, M IP _Ly and the corresponding optimal channel fingerprint feature combination rule f IP _Lx, f IP The feature selection module includes the following sub-steps:

[0155] Step S401: After inputting the channel fingerprint features and the corresponding horizontal (vertical) coordinates x (y) marked as line-of-sight labels in the training set into RF for training, the training model gives the channel fingerprint feature importance score;

[0156] Step S402: sort the channel fingerprint features of the training set in descending order according to the importance scores of the channel fingerprint features, and record the sorting rule f;

[0157] Step S403: For the sorted training set channel fingerprint features, select the first n channel fingerprint features each time and record the sorting rule f of the channel fingerprint features. n, the first n channel fingerprint features selected together with the corresponding horizontal (vertical) coordinates x (y) are input into RF for training and n training models M are generated x (M y ), the channel fingerprint features marked as line-of-sight labels in the verification set are pressed by f n After reordering, input the nth training model M x (M y ), estimate the horizontal (vertical) coordinate x(y) of the validation set under n training models;

[0158] Step S404: Combine the estimated horizontal (vertical) coordinates x (y) of the validation set into complete coordinates, and calculate the relative error between the coordinates and the true validation set coordinates. Then calculate the average relative error of the n training models, and take the training model with the lowest average relative error as M. IP _Lx, M IP _Ly, the channel fingerprint feature sorting rule corresponding to this model is denoted as f IP _Lx, f IP _Ly.

[0159] Step S5: The horizontal coordinate x and the corresponding channel fingerprint feature, the vertical coordinate y and the corresponding channel fingerprint feature of the offline reference point marked with the non-line-of-sight label in the training set are input into RF for training. After executing the feature selection module, the positioning model M is output. IP _Nx, M IP _Ny and the corresponding optimal channel fingerprint feature combination rule f IP _Nx, f IP _Ny; the feature selection module includes the following sub-steps:

[0160] Step S501: After inputting the channel fingerprint features and the corresponding horizontal (vertical) coordinates x (y) of the training set marked as non-line-of-sight labels into RF for training, the training model gives the channel fingerprint feature importance score;

[0161] Step S502: sort the channel fingerprint features of the training set in descending order according to the importance scores of the channel fingerprint features, and record the sorting rule f;

[0162] Step S503: For the sorted training set channel fingerprint features, select the first n channel fingerprint features each time and record the sorting rule f of the channel fingerprint features. n , the first n channel fingerprint features selected together with the corresponding horizontal (vertical) coordinates x (y) are input into RF for training and n training models M are generated x (M y ), the channel fingerprint features marked as non-line-of-sight labels in the verification set are pressed by f n After reordering, input the nth training model Mx (M y ), estimate the horizontal (vertical) coordinate x(y) of the validation set under n training models;

[0163] Step S504: Combine the estimated x and y coordinates of the validation set into a complete coordinate, and calculate the relative error between the coordinate and the true validation set coordinate. Then calculate the average relative error of the n training models, and take the training model with the lowest average relative error as M. IP _Nx, M IP _Ny, the channel fingerprint feature sorting rule corresponding to this model is denoted as f IP _Nx, f IP _Ny.

[0164] Step S6: Collect channel state information for the target location point and extract the channel fingerprint features of the target location point based on the novel channel fingerprint structure. This acquisition of channel state information for the offline reference point can utilize ray tracing-based wireless channel simulation software or wireless channel measurement equipment. In this example, the acquisition of channel state information for the offline reference point utilizes both the commercial ray tracing simulation software Wireless InSite and wireless channel measurement equipment. The Wireless InSite process specifically includes the following sub-steps:

[0165] Step S601: Set the height of the offline reference point to 1.5m, the height of the transmitter to 2.8m, both the transmitter and the receiver use omnidirectional antennas, the transmitter is a 4×4 array antenna, the receiver is a 2×2 array antenna, the transmitter position coordinates are (13.59m, 14.11m), 200 target positioning points are randomly arranged in the positioning scene, the electromagnetic wave frequency is 60GHz, the bandwidth is 3GHz, the transmission power is 23dBm, the maximum allowable reflection order is 6, and the diffraction order is 1;

[0166] Step S602: In Wireless InSite, output the channel simulation results and each target location can be marked with a line-of-sight / non-line-of-sight label. Each ray in the simulation results corresponds to a set of channel state information, including received signal strength, azimuth arrival angle, elevation arrival angle, and arrival time.

[0167] Collecting the channel state information of the target positioning point using wireless channel measurement equipment includes the following sub-steps:

[0168] Step S603: Set the transmitting and receiving antennas to 4×8 dual-polarized cylindrical MIMO and 4×4 dual-polarized planar MIMO, with a bandwidth of 700 MHz. After calibrating the wireless channel measurement equipment, perform channel measurements on each offline reference point and record the offline reference point coordinates and line-of-sight / non-line-of-sight labels. Collect channel state information for a total of 130 offline reference points, including 90 reference points in the aisle, representing a line-of-sight environment, and 30 reference points on the table, representing a non-line-of-sight environment. Allocate 20% of the reference points as target positioning points.

[0169] Step S604: Use the spatial alternating generalized expectation algorithm to perform channel estimation on the channel measurement results to obtain channel state information for each subcarrier, including: received signal strength, azimuth arrival angle, elevation arrival angle, and arrival time;

[0170] The novel channel fingerprint structure includes the following sub-steps:

[0171] Step S605: reconstructing the delay / angular power spectrum density based on the channel state information in step S602 or step S604;

[0172] Step S606: Extract channel fingerprint features based on the MRP path, including received signal strength, azimuth arrival angle, elevation arrival angle, and arrival time;

[0173] Step S607: extract channel characteristics based on the reconstructed delay / angular power spectrum density, including: received power kurtosis, received power skewness, rise time, root mean square delay spread, root mean square angular spread and Ricean K factor.

[0174] Step S7: The channel fingerprint characteristics of the target positioning point are combined according to the optimal channel fingerprint characteristic combination rule f I After reorganization and permutation, input line-of-sight / non-line-of-sight classification model M I , estimate the line-of-sight / non-line-of-sight label of the target positioning point;

[0175] Step S8, according to the estimated line-of-sight / non-line-of-sight label of the target positioning point, select the corresponding positioning model and the corresponding optimal channel fingerprint feature combination rule, re-arrange the channel fingerprint features of the target positioning point according to the optimal channel fingerprint feature combination, input the positioning model, and perform adaptive K value weighted prediction, and finally estimate the position coordinates of the target positioning point. The specific process of selecting the corresponding positioning model and the corresponding optimal channel fingerprint feature combination according to the estimated line-of-sight / non-line-of-sight label of the target positioning point is as follows: if the estimated line-of-sight / non-line-of-sight label of the target positioning point is 1, then select the training model M IP _Lx, M IP _Ly, and the corresponding channel fingerprint feature sorting rule f IP _Lx, f IP_Ly, if the estimated line-of-sight / non-line-of-sight label of the target positioning point is 0, then the training model M is selected IP _Nx, M IP _Ny, and the corresponding channel fingerprint feature sorting rule f IP _Nx, f IP _Ny.

[0176] The adaptive K value weighted prediction, the specific steps are:

[0177] Step S801: After the channel fingerprint features of the target positioning point are rearranged according to the optimal channel fingerprint feature combination and input into the selected positioning model, the scores score_x and score_y of the horizontal and vertical coordinates of the offline reference point in the channel fingerprint database can be obtained;

[0178] Step S802: Use K-means clustering algorithm to accurately divide the scores score_x and score_y into two categories, one with higher scores and one with lower scores. The number of coordinates in the category with higher scores is the adaptive K value K. x , K y ;

[0179] Step S803: The K with the highest score x The horizontal axis x k and its score score_x k , K y The vertical coordinate y k and its score score_y k , respectively with the horizontal coordinate score score_x k , vertical coordinate score score_y k is the horizontal coordinate x k , vertical coordinate y k The weight of the target positioning point is obtained by calculating the final prediction coordinates (x T ,y T ):

[0180]

[0181] Where k is K x , K y The index of .

[0182] like Figure 9As shown in the figure, the ray tracing simulation data verifies the superior performance of the proposed line-of-sight / non-line-of-sight recognition assisted positioning method suitable for 6G indoor communications, and has the fastest error convergence speed. The error of the proposed method is within 1m with a probability of 95%. Compared with BiLSTM, the performance of the proposed method is significantly enhanced, and the average positioning error is improved by about 45%. Through weighted prediction and adaptive K value, the positioning performance of weighted random forest (WRF) is 30% higher than that of RF. With the help of feature selection and line-of-sight / non-line-of-sight recognition modules, the positioning performance of IWRF is 17% higher than that of WRF, which proves the effectiveness of feature selection and line-of-sight / non-line-of-sight recognition. Feature selection reduces redundant information and prevents mutual interference between channel fingerprint features, ultimately improving positioning accuracy. Furthermore, in a measurement environment with more interference factors, the performance of the proposed method still remains optimal, which proves the robustness of the method proposed in the present invention, such as Figure 10 As shown in the figure, among the compared algorithms, the proposed method exhibits the fastest error convergence and the highest degree of error concentration, indicating its potential application value for practical indoor positioning. Therefore, the proposed line-of-sight / non-line-of-sight identification-assisted positioning method for 6G indoor communications has high accuracy, strong stability, and greater versatility, and can meet the centimeter-level positioning accuracy required by 6G systems.

[0183] The line-of-sight / non-line-of-sight recognition assisted positioning method for 6G indoor communications according to an embodiment of the present invention extracts channel fingerprint features for offline reference points; inputs the channel fingerprint feature set of the offline reference points into the optimized weighted random forest algorithm to perform line-of-sight / non-line-of-sight recognition and positioning training respectively, and outputs recognition and positioning models respectively; extracts channel fingerprint features for the target positioning point; inputs the channel fingerprint features of the target positioning point into the line-of-sight / non-line-of-sight recognition model, and inputs the channel fingerprint features of the target positioning point into the corresponding positioning model based on the predicted line-of-sight / non-line-of-sight label, and outputs the predicted coordinates. The present invention can solve the problem of low positioning accuracy caused by multipath fading and non-line-of-sight propagation in indoor environments, greatly improves the accuracy and robustness of positioning, and is superior to existing algorithms. In simulations and actual measurements of different scenarios, this method has demonstrated optimal performance and has potential applicability even without a large amount of training data.

[0184] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or N embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and the features of different embodiments or examples without contradiction.

[0185] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of technical features indicated. Thus, a feature specified as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of the present invention, "N" means at least two, such as two, three, etc., unless otherwise specifically defined.

[0186] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code comprising one or N executable instructions for implementing a custom logical function or step of a process, and the scope of the preferred embodiments of the present invention includes alternative implementations in which functions may be performed out of the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present invention pertain.

Claims

1. A line-of-sight / non-line-of-sight identification assisted positioning method suitable for 6G indoor communication, characterized in that: The following steps are involved: Collecting channel state information of the target positioning point, and extracting a channel fingerprint feature of the target positioning point according to the channel state information; Inputting the channel fingerprint features of the target positioning point into a pre-built line-of-sight / non-line-of-sight classification model according to the optimal channel fingerprint feature combination rule, and outputting the line-of-sight / non-line-of-sight label of the target positioning point; According to the line-of-sight / non-line-of-sight label of the target positioning point, a pre-built positioning model and the corresponding optimal channel fingerprint feature combination rule are selected. The channel fingerprint features of the target positioning point are recombined and arranged according to the optimal channel fingerprint feature combination rule, input into the positioning model, and adaptive K value weighted prediction is performed to estimate the position coordinates of the target positioning point.

2. The method according to claim 1, characterized in that The channel state information of the target positioning point is collected using wireless channel simulation software or wireless channel measurement equipment based on ray tracing, wherein the wireless channel simulation software based on ray tracing collects the channel state information of the target positioning point, including: In the ray tracing-based wireless channel simulation software, an indoor scene is constructed and simulation parameters are set. The channel simulation results are output, and each target location can be marked with a line-of-sight or non-line-of-sight label. Each ray in the simulation results corresponds to a set of channel state information, including received signal strength, azimuth arrival angle, elevation arrival angle, and arrival time. The channel state information of the target positioning point is collected using a wireless channel measurement device, including: calibrating the wireless channel measurement device, performing channel measurement on the target positioning point and recording the coordinates of the target positioning point and the line-of-sight / non-line-of-sight label; and using a spatial alternating generalized expectation algorithm to perform channel estimation on the channel measurement results to obtain channel state information of each subcarrier.

3. The method according to claim 2, characterized in that Extracting a channel fingerprint feature of the target positioning point according to the channel state information includes: Reconstruct the delay / angular power spectrum density based on the collected channel state information; Extracting channel fingerprint features based on the normalized maximum power path, wherein the channel fingerprint features include: received signal strength, azimuth arrival angle, elevation arrival angle and arrival time; Channel characteristics are extracted based on the reconstructed delay / angular power spectral density, including received power kurtosis, received power skewness, rise time, RMS delay spread, RMS angular spread, and Ricean K factor.

4. The method according to claim 1, wherein Performing adaptive K-value weighted prediction to estimate the position coordinates of the target positioning point includes: After the channel fingerprint features of the target positioning point are rearranged according to the optimal channel fingerprint feature combination and input into the selected positioning model, the scores score_x and score_y of the horizontal and vertical coordinates of the offline reference point in the channel fingerprint database are obtained; The K-means clustering algorithm is used to divide the scores score_x and score_y into high-level categories and low-level categories. The number of coordinates contained in the high-level category is the adaptive K value K. x , K y ; The K with the highest score x The horizontal axis x k and its score score_x k , K y The vertical coordinate y k and its score score_y k , respectively with the horizontal coordinate score score_x k , vertical coordinate score score_y k is the horizontal coordinate x k , vertical coordinate y k The weight of the target positioning point is obtained by calculating the final prediction coordinates (x T ,y T ): Where k is K x , K y The index of .

5. The method according to claim 2, characterized in that The method further includes: constructing a line-of-sight / non-line-of-sight recognition model, specifically comprising the following steps: Establish a reference grid within the positioning area, with the grid lines perpendicular to the coordinate axes of the Cartesian coordinate system. Take the intersection of the grid points as offline reference points, mark the offline reference points with line-of-sight / non-line-of-sight labels, and collect channel status information for the offline reference points. Extracting channel fingerprint features of offline reference points based on the channel fingerprint structure, building a channel fingerprint database based on the channel fingerprint features, offline reference point coordinates, and line-of-sight / non-line-of-sight labels, and dividing the channel fingerprint database into a training set and a validation set; The channel fingerprint features and line-of-sight / non-line-of-sight labels of the training set are input into the random forest model for training. After executing the feature selection module, the line-of-sight / non-line-of-sight classification model M is output. I And the corresponding optimal channel fingerprint feature combination rule f I ; The horizontal coordinate x and the corresponding channel fingerprint feature, the vertical coordinate y and the corresponding channel fingerprint feature of the offline reference point marked as the line-of-sight label in the training set are input into the random forest model for training. After executing the feature selection module, the corresponding positioning model M is output. IP _Lx, M IP _Ly and the corresponding optimal channel fingerprint feature combination rule f IP _Lx, f IP _Ly; The horizontal coordinate x and the corresponding channel fingerprint feature, the vertical coordinate y and the corresponding channel fingerprint feature of the offline reference point marked as non-line-of-sight label in the training set are input into the random forest model for training. After executing the feature selection module, the positioning model M is output. IP _Nx, M IP _Ny and the corresponding optimal channel fingerprint feature combination rule f IP _Nx, f IP _Ny.

6. The method according to claim 5, characterized in that Label offline reference points with line-of-sight / non-line-of-sight labels, including: If the transmission between the offline reference point and the transmitter is in line of sight, the line of sight / non-line of sight tag of the offline reference point is marked as 1. If the transmission between the offline reference point and the transmitter is in non-line of sight, the line of sight / non-line of sight tag of the offline reference point is marked as 0.

7. The method according to claim 5, characterized in that The channel state information of the offline reference point is collected using wireless channel simulation software or wireless channel measurement equipment based on ray tracing, wherein the wireless channel simulation software based on ray tracing collects the channel state information of the offline reference point, including: In the ray tracing-based wireless channel simulation software, an indoor scenario is constructed and simulation parameters are set. The channel simulation results are output, along with line-of-sight / non-line-of-sight labels for each offline reference point. Each ray in the simulation results corresponds to a set of channel state information, including received signal strength, azimuth arrival angle, elevation arrival angle, and arrival time. Collecting channel state information of offline reference points using wireless channel measurement equipment includes: calibrating the wireless channel measurement equipment, performing channel measurement on each offline reference point and recording the coordinates and line-of-sight / non-line-of-sight labels of the offline reference points; and performing channel estimation on the channel measurement results using a spatial alternating generalized expectation algorithm to obtain channel state information of each subcarrier.

8. The method according to claim 5, characterized in that The channel fingerprint features and line-of-sight / non-line-of-sight labels of the training set are input into the random forest model for training. After executing the feature selection module, the line-of-sight / non-line-of-sight classification model M is output. I And the corresponding optimal channel fingerprint feature combination rule f I ,include: After the channel fingerprint features and line-of-sight / non-line-of-sight labels of the training set are input into the random forest model for training, the random forest model gives the channel fingerprint feature importance score; According to the importance score of the channel fingerprint feature, sort the channel fingerprint features of the training set in descending order and record the sorting rule f; For the channel fingerprint features of the sorted training set, select the first n channel fingerprint features each time and record the sorting rule f of the channel fingerprint features n , input the first n channel fingerprint features selected together with the line-of-sight / non-line-of-sight labels into the random forest model for training and generate n training models, and the channel fingerprint features of the validation set are calculated according to f n After reordering, input the nth training model respectively to estimate the line-of-sight / non-line-of-sight labels of the validation set under the n training models; Calculate the consistency ratio of the line-of-sight / non-line-of-sight labels of the n groups of estimated validation sets and the line-of-sight / non-line-of-sight labels of the real validation sets, and get the accuracy of the n training models. Take the training model with the highest accuracy as the line-of-sight / non-line-of-sight classification model M I , line-of-sight / non-line-of-sight classification model M I The corresponding channel fingerprint feature sorting rule is denoted as f I .

9. The method according to claim 5, characterized in that The horizontal coordinate x and the corresponding channel fingerprint feature, the vertical coordinate y and the corresponding channel fingerprint feature of the offline reference point marked as the line-of-sight label in the training set are input into the random forest model for training. After executing the feature selection module, the corresponding positioning model M is output. IP _Lx, M IP _Ly and the corresponding optimal channel fingerprint feature combination rule f IP _Lx, f IP _Ly, including: After inputting the channel fingerprint features marked as line-of-sight labels in the training set and the corresponding horizontal coordinates x and vertical coordinates y into the random forest model for training, the training model gives the channel fingerprint feature importance score; According to the importance score of the channel fingerprint feature, sort the channel fingerprint features of the training set in descending order and record the sorting rule f; For the sorted training set channel fingerprint features, select the first n channel fingerprint features each time and record the sorting rule f of the channel fingerprint features n , input the first n channel fingerprint features selected together with the corresponding horizontal coordinate x / vertical coordinate y into the random forest model for training and generate n training models M x / M y , the channel fingerprint features marked as line-of-sight labels in the verification set are pressed by f n After reordering, input the nth training model M x / M y , estimate the horizontal coordinate x / vertical coordinate y of the validation set under n training models; Combine the estimated x-coordinate and y-coordinate of the validation set into complete coordinates, calculate the absolute error between the estimated complete coordinates and the true validation set coordinates, calculate the average absolute error of the n training models, and take the training model with the lowest average absolute error as the positioning model M IP _Lx, M IP _Ly, the channel fingerprint feature sorting rule corresponding to the positioning model is recorded as f IP _Lx, f IP _Ly.

10. The method according to claim 5, characterized in that The horizontal coordinate x and the corresponding channel fingerprint feature, the vertical coordinate y and the corresponding channel fingerprint feature of the offline reference point marked as non-line-of-sight label in the training set are input into the random forest model for training. After executing the feature selection module, the positioning model M is output. IP _Nx, M IP _Ny and the corresponding optimal channel fingerprint feature combination rule f IP _Nx, f IP _Ny, including: After inputting the channel fingerprint features and the corresponding x-coordinates / y-coordinates of the training set marked as non-line-of-sight into the random forest model for training, the training model gives the importance score of the channel fingerprint features. According to the importance score of the channel fingerprint feature, sort the channel fingerprint features of the training set in descending order and record the sorting rule f; For the sorted training set channel fingerprint features, select the first n channel fingerprint features each time and record the sorting rule f of the channel fingerprint features n , input the first n channel fingerprint features selected together with the corresponding horizontal coordinate x / vertical coordinate y into the random forest model for training and generate n training models M x / M y , the channel fingerprint features marked as non-line-of-sight labels in the verification set are pressed by f n After reordering, input the nth training model M x / M y , estimate the horizontal coordinate x / vertical coordinate y of the validation set under n training models; Combine the estimated horizontal coordinate x and vertical coordinate y of the validation set into the complete coordinate, and calculate the absolute error between the estimated complete coordinate and the true validation set coordinate. Then calculate the average absolute error of n training models, and take the training model with the lowest average absolute error as the positioning model M IP _Nx, M IP _Ny, the channel fingerprint feature sorting rule corresponding to the positioning model is recorded as f IP _Nx, f IP _Ny.

Images