A non-line-of-sight signal recognition method based on channel feature weighted model

Abstract

The present invention relates to a non-line-of-sight signal recognition method based on a channel feature weighted model. The method comprises the following steps: after acquiring channel impulse response (CIR) data, nine features are constructed on the CIR data. First, seven basic channel features are extracted from a CIR waveform diagram. Two new variables, an attenuation factor (Mr) and a peak time factor (IDiff), are further constructed based on signal propagation characteristics. The attenuation factor (Mr) quantifies the attenuation degree of a first path, and the peak time factor (IDiff) measures the difference between the first path and the peak path position under different channels, fully reflecting the differences between the two variables between different channels. The method further calculates the weight of each feature according to a relief algorithm, explores the ability of the feature to distinguish close-range samples, introduces weight coefficients, and performs support vector machine (SVM) classification on the weighted features to improve the ability to recognize non-line-of-sight signals.

Description

Technical Field

[0001] The invention belongs to the technical field of wireless positioning and relates to a non-line-of-sight signal recognition method based on a feature weighted model. Background Art

[0002] With the rapid development of the national economy, positioning technology has matured. Satellite navigation-based services are widely used in outdoor environments. In recent years, location-based service technologies have also been evolving towards indoor positioning systems. With the continuous advancement of modern society and the increasing number of large buildings, people spend 70% of their time indoors, resulting in a growing demand for indoor location-based services. For example, accurate indoor location information is required for locating special populations, office space management, and logistics warehousing. High-precision indoor positioning holds enormous potential for application in emergency scenarios, such as firefighting, public safety, and earthquake relief. Among current wireless positioning technologies, ultra-wideband positioning technology, unlike traditional communication technologies, does not utilize the carrier wave of conventional communication systems. Instead, it transmits data by sending and receiving extremely narrow pulses in the nanosecond or sub-nanosecond range, making it a leading indoor positioning technology.

[0003] In ultra-wideband indoor positioning systems, various obstacles indoors can block ultra-wideband signals, causing non-line-of-sight (NLOS) propagation. This NLOS propagation reduces the reliability of communication and positioning accuracy. Therefore, accurately identifying NLOS signals during positioning and taking appropriate measures can improve positioning accuracy. Traditional NLOS detection techniques can be categorized into three types: distance estimation-based methods, channel impulse response-based methods, and position estimation-based methods. Distance-based methods primarily distinguish between line-of-sight (LOS) and non-line-of-sight (NLOS) using probability density functions or the variance of range estimates. While seemingly simple, these methods often suffer from poor positioning performance due to limitations in prior distribution functions or time delays. Position-based methods identify NLOS during the position estimation process, and can even use the calculated position coordinates sequentially for NLOS detection. These methods can detect NLOS by comparing position estimates generated from different subsets of distance estimates when redundant distance estimates are available, but are ineffective when redundant distance estimates are unavailable. The method based on channel impulse response is widely used for NLOS signal identification because it can make full use of the statistical information of the received multipath components and use channel parameters to determine whether the received signal is an NLOS signal or a LOS signal. Summary of the Invention

[0004] To address the problem of traditional methods based on channel impulse response (CIR) failing to consider the importance of information contained in channel features, this paper proposes a non-line-of-sight (NLOS) signal recognition method based on a weighted channel feature model. This method constructs two new feature variables, Mc and IDiff, based on the channel impulse response (CIR) and signal propagation characteristics. This method combines seven basic channel features to form a feature set consisting of nine feature variables. The method uses a relief algorithm to calculate the weights of each feature, fully exploiting the feature's ability to distinguish close-range samples. Weight coefficients are introduced to weight the features for support vector machine (SVM) classification, thereby improving NLOS signal recognition.

[0005] In order to achieve the above object, the present invention provides a non-line-of-sight signal recognition method based on a channel feature weighted model, comprising the following steps:

[0006] (1) Obtaining training samples: Collect N channel impulse response samples with labeled signal categories as the training set S = {(x1(t), y1), (x2(t), y2), ..., (x N (t),y N )}; where x n (t)(n=1,2,…,N) represents the nth sample in the training set, which is a 1×T vector, where T represents the time dimension of the sample; y n Represents sample x n (t) label, y n ∈{0,1};

[0007] (II) Feature construction: 9 channel features are constructed from the training set samples, of which 7 basic channel features are: total energy (ε), maximum amplitude (r max ), rise time (t rise ), kurtosis (κ), average additional delay (t med ), RMS delay spread (t rms ) and skewness (ske); and construct two new characteristic parameters, attenuation factor (Mc) and peak time factor (IDiff); the attenuation factor (Mc) quantifies the attenuation degree of the first path, and the peak time factor (IDiff) measures the difference between the first path and the peak path position under different channels; thus constructing the feature set F = {ε,r max ,t rise ,κ,t med ,t rms ,ske,Mc,IDiff};

[0008] (3) Feature preprocessing: normalize each channel feature;

[0009] (4) Establish a feature-weighted non-line-of-sight signal recognition model: Use the Relief algorithm to calculate the weights of each feature in the feature set F, form a feature weight vector w, and weight the corresponding elements in the feature set F to obtain a weighted feature set F' = F·diag(w); Use the Support Vector Machine (SVM) to train the training set S, where the input of the SVM is the new feature set F' weighted by the Relief algorithm to solve the problem that directly using the original feature set F ignores potential key information;

[0010] (5) Generating non-line-of-sight signal recognition results: placing the test sample into a non-line-of-sight signal recognition model based on channel feature weighting to obtain the final non-line-of-sight signal recognition results.

[0011] Furthermore, in the step (1), N channel impulse response samples with signal category labels are collected as a training set S = {(x1(t), y1), (x2(t), y2), ..., (x N (t),y N )},x n (t)(n=1,2,…,N) represents the nth sample in the training set, which is a 1×T vector, where T represents the time dimension of the sample; y n Represents sample x n (t) label, y n ∈{0,1};

[0012] Furthermore, in the step (2), features are constructed and extracted from the training set samples. First, 7 basic channel features are calculated. For a general training sample x(t) (t=t1, t2, ... t T ), the mathematical model of its characteristic parameters is expressed as follows:

[0013] (1) Total Energy

[0014]

[0015] (2) Maximum Amplitude

[0016] r max =max|x(t i )| (2)

[0017] (3) Rise Time

[0018] t rise =t H -t L (3)

[0019] Among them, t HIndicates the time required for the signal to reach 0.9 times the maximum amplitude, t L Indicates the time required for the signal to reach 0.1 times the maximum amplitude;

[0020] (4) Kurtosis

[0021]

[0022] (5) Mean Excess Delay

[0023]

[0024] (6) Root Mean Square Delay Spread

[0025]

[0026] (7) Skewness

[0027]

[0028] Among them, x(t i ) indicates that the sample x(t) is at t=t i The value at the moment, μ x represents the mean of the sample x(t), σ x represents the standard deviation of the sample x(t);

[0029] Furthermore, two new characteristic parameters, the attenuation factor (Mc) and the peak time factor (IDiff), are constructed as follows:

[0030] (1) Attenuation factor (Mc)

[0031]

[0032] Where fp max =max{fp1,fp2,fp3}, where fp1, fp2 and fp3 represent the amplitudes of the three most adjacent signals after the first path;

[0033] (2) Peak time factor (IDiff)

[0034] IDiff=|t fp -t pp | (9)

[0035] Where, t fp is the path position of the first path, t pp is the path position of the peak;

[0036] The attenuation factor (Mr) quantifies the attenuation of the first path amplitude. A larger value indicates that the channels are more likely to be in line-of-sight.

[0037] The peak time factor (IDiff) measures the difference between the first path and the peak path position in different channels. Typically, in line-of-sight (LOS) conditions, the difference between the first path and the peak path position is much closer than in non-line-of-sight (NLOS) conditions. Therefore, this metric is used to determine the likelihood of non-line-of-sight conditions.

[0038] Thus, a feature set F={ε,r max ,t rise ,κ,t med ,t rms ,ske,Mc,IDiff};

[0039] Furthermore, in step (3), each feature extracted based on the channel impulse response is normalized, and the normalization formula is:

[0040]

[0041] Where X is the original feature data, X min is the minimum value in the original feature data, X max is the maximum value in the original feature data;

[0042] Furthermore, in step (4), first for the j-th feature F j ∈R N×1 (where j = 1 to 9), use the Relief algorithm to calculate the weight w(j) of each feature, and obtain the feature weight vector w = [w(1), w(2), ..., w(9)]. The calculation formula for each feature weight is as follows:

[0043]

[0044] In the formula, the initial weight of w(j) is 0, w k (j) is the weight of the jth feature in the feature set F in the kth update, f∈F j ,diff(F j ,f,H) and diff(F j ,f,M) respectively select the feature element f and its similar nearest neighbor sample H and different class nearest neighbor sample M in feature F j The difference between the two, m is the number of sample sampling;

[0045] Furthermore, the feature weight vector w is multiplied by the corresponding element in the training set to obtain the weighted feature set F' = F·diag(w); the training set is trained using Support Vector Machine (SVM), where the input of the SVM is the new feature set F' weighted by the Relief algorithm. The weight coefficient is introduced to fully explore the importance of different features to the classification model;

[0046] Furthermore, in step (5), the test sample is put into a non-line-of-sight signal recognition model based on channel feature weighting to obtain the final recognition result.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] The present invention provides a non-line-of-sight signal recognition method based on a channel feature weighted model. Two new characteristic variables, an attenuation factor (Mc) and a peak time factor (IDiff), are constructed based on the channel impulse response (CIR) and signal propagation characteristics. The method quantifies the attenuation degree of the first path and measures the difference between the first path and the peak path position under different channels. This method fully exploits the information contained in different channels and combines basic channel features to form a feature set consisting of nine characteristic variables. The method further considers the importance of features to the classification model, calculates the weight of each feature using a relief algorithm, introduces weight coefficients based on the feature's ability to distinguish close-range samples, and performs support vector machine (SVM) classification on the weighted features to achieve improved non-line-of-sight signal recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 Flowchart of the non-line-of-sight signal recognition method based on the channel feature weighted model of the present invention;

[0050] Figure 2 This is a distribution diagram of channel characteristic parameters under LOS and NLOS conditions according to an embodiment of the present invention; DETAILED DESCRIPTION

[0051] The present invention is described in detail below by way of exemplary embodiments, but it should be understood that elements, structures, and features of one embodiment may be beneficially combined in other embodiments without further description.

[0052] See also Figure 1 The present invention discloses a non-line-of-sight signal recognition method based on a channel feature weighted model, comprising the following steps:

[0053] (1) Obtaining training samples: Collect N channel impulse response samples with labeled signal categories as the training set S = {(x1(t), y1), (x2(t), y2), ..., (x N (t),y N)}; where x n (t)(n=1,2,…,N) represents the nth sample in the training set, which is a 1×T vector, where T represents the time dimension of the sample; y n Represents sample x n (t) label, y n ∈{0,1}; For example, the DWM1000 module can be used for data acquisition, with a pulse repetition frequency of 64 MHz, and each data has 1016 samples, and the time resolution is about half a period of the 499.2 MHz fundamental frequency;

[0054] (II) Feature Construction: Construct 9 channel features from the training set samples, and first calculate 7 basic channel features; for a general training sample x(t) (t=t1,t2,…t T ), the mathematical model of its characteristic parameters is expressed as follows:

[0055] (1) Total Energy

[0056]

[0057] (2) Maximum Amplitude

[0058] r max =max|x(t i )| (13)

[0059] (3) Rise Time

[0060] t rise =t H -t L (14)

[0061] Among them, t H Indicates the time required for the signal amplitude to reach 0.9 times the maximum amplitude, t L Indicates the time required for the signal amplitude to reach 0.1 times the maximum amplitude;

[0062] (4) Kurtosis

[0063]

[0064] (5) Mean Excess Delay

[0065]

[0066] (6) Root Mean Square Delay Spread

[0067]

[0068] (7) Skewness

[0069]

[0070] Among them, x(t i ) indicates that the sample x(t) is at t=t i The value at the moment, μ x represents the mean of the sample x(t), σ x| represents the standard deviation of the sample x(t);

[0071] Furthermore, two new characteristic parameters, the attenuation factor (Mc) and the peak time factor (IDiff), are constructed as follows:

[0072] (1) Attenuation factor (Mc)

[0073]

[0074] Where fp max =max{fp1,fp2,fp3}, where fp1, fp2 and fp3 represent the amplitudes of the three nearest neighbor signals after the first path;

[0075] (2) Peak time factor (IDiff)

[0076] IDiff=|t fp -t pp | (20)

[0077] Where, t fp is the path position of the first path, t pp is the path position of the peak;

[0078] Thus, a feature set F={ε,r max ,t rise ,κ,t med ,t rms ,ske,Mc,IDiff};

[0079] (3) Feature preprocessing: Normalize each feature extracted based on the channel impulse response. The normalization formula is:

[0080]

[0081] Where X is the original feature data, X min is the minimum value in the original feature data, X max is the maximum value in the original feature data;

[0082] (IV) Establishing a feature-weighted non-line-of-sight signal recognition model: First, for the j-th feature F j ∈R N×1 (where j = 1 to 9), use the Relief algorithm to calculate the weight w(j) of each feature, and obtain the feature weight vector w = [w(1), w(2), ..., w(9)]. The calculation formula for each feature weight is as follows:

[0083]

[0084] In the formula, the initial weight of w(j) is 0, w k (j) is the weight of the jth feature in the feature set F in the kth update, f∈F j ,diff(F j ,f,H) and diff(F j ,f,M) respectively select the feature element f and its similar nearest neighbor sample H and different class nearest neighbor sample M in feature F j The difference between the two, m is the number of sample sampling;

[0085] Furthermore, the feature weight vector w is multiplied by the corresponding element in the training set to obtain the weighted feature set F' = F·diag(w); the training set is trained using Support Vector Machine (SVM), where the input of the SVM is the new feature set F' weighted by the Relief algorithm. The weight coefficient is introduced to fully explore the importance of different features to the classification model;

[0086] (5) Generating non-line-of-sight signal recognition results: placing the test sample into a non-line-of-sight signal recognition model based on channel feature weighting to obtain the final recognition result.

[0087] The non-line-of-sight signal recognition method of the present invention constructs nine channel features. First, seven basic channel features are extracted from the channel impulse response waveform. Further, two new characteristic variables, the attenuation factor (Mr) and the peak time factor (IDiff), are constructed based on the signal propagation characteristics. The attenuation degree of the first path is quantified. At the same time, a measurement is made for the difference between the first path and the peak path position under different channels. The information contained in different channels is fully exploited, and the basic channel features are combined to form a feature set consisting of nine characteristic variables. At the same time, the importance of features to the classification model is further considered. The weight of each feature is further calculated using the Relief algorithm, and the features are weighted as the input of the SVM classifier to construct a classification model based on the weighted channel features. The test sample is placed in the constructed classification model to obtain the non-line-of-sight signal recognition result.

[0088] The non-line-of-sight signal recognition method of the present invention constructs two new feature variables and introduces a relief feature weighting strategy, thereby improving the performance of recognizing non-line-of-sight signals.

[0089] In order to more clearly illustrate the beneficial effects of the non-line-of-sight signal recognition method of the present invention, the non-line-of-sight signal recognition method of the present invention is further described below with reference to embodiments.

[0090] Example:

[0091] The data used in this example comes from the elastic Wireless Networking Experimentation (eWINE), a subproject of the EU's Horizon 2020 research program. The program leverages artificial intelligence to address current challenges in dense and dynamic wireless networks. The data was measured using a DW1000 ultra-wideband transceiver chip, configured with four channels (500MHz bandwidth, 3993.6MHz center frequency), a data length of 1024, and a data rate of 110kb / s. The dataset contains 42,000 samples across seven different measurement scenarios: Office Environment 1, Office Environment 2, Apartment, Workshop, Living Room, Bedroom, and Boiler Room. To prevent overfitting in specific environments, strict node positioning was not performed during the measurement process. Instead, a UWB node was placed at a random fixed location within the measurement environment, and measurements were performed while the other UWB node was moved throughout the environment. For each environment, 3,000 measurements were collected under line-of-sight (LOS) channel conditions, followed by 3,000 measurements under non-line-of-sight (NLOS) channel conditions within the same environment. This example analyzes 6000 samples from the dataset, including 4800 training data and 1200 test data. Each row in the dataset is 1024 characters long and contains 1016 channel impulse response data (CIR0-CIR1015) along with parameters such as whether the signal is an NLOS signal.

[0092] First, the basic channel features of the CIR waveform signal are extracted according to step (2), and 7 basic channel features are extracted. Two new features Mr and IDiff are constructed based on the propagation characteristics of the signal. All the features are obtained. The channel characteristic parameter distribution diagram under LOS and NLOS conditions is shown as follows: Figure 2As shown in the figure, it can be seen that the overlap of skewness (ske) and kurtosis (κ) is relatively high, and classification based on numerical values ​​alone is not feasible. The overlap of the remaining feature parameters is low, and they can be used in classification, but the classification effect of using a single feature is not good. Therefore, it is impossible to judge which features to use as the basis for classification by simply observing the distribution values ​​of several feature parameters. Taking this problem into consideration, this patent introduces a feature weight coefficient and uses the Relief algorithm. Based on the feature's ability to distinguish close-range samples, different weights are assigned to features according to the correlation between each feature and the category, fully exploring the importance of different features to the classification model. The features are weighted to establish a Relief-SVM classification model to improve the ability to recognize non-line-of-sight signals.

[0093] In this example, Method 1: The impulse response data CIR is not subjected to basic channel feature extraction and used as the input of the SVM classifier; Method 2: The impulse response data CIR is subjected to basic channel feature extraction, and 7 basic channel features are extracted as the input of the SVM classifier; Method 3: The impulse response data CIR is subjected to basic channel feature extraction and new feature parameters are constructed, and 9 basic channel features are extracted as the input of the SVM classifier; Method 4: The method proposed in this patent is used to perform non-line-of-sight signal recognition, and its recognition accuracy is shown in Table 1.

[0094] Table 1

[0095] method Dharma One Method 2 Method 3 The present invention Recognition rate 74.75％ 76.75％ 80.08％ 81.17％

[0096] As can be seen from Table 1, the method provided by the present invention achieves the best recognition result and has the best non-line-of-sight signal recognition performance.

[0097] Based on the above analysis, the method provided by the present invention constructs new channel features and introduces relief weight coefficients. It assigns different weights to features according to their relevance to categories, explores the importance of different features to the classification model, and weights the channel features. It has a good non-line-of-sight signal recognition capability.

[0098] The above embodiments are only used to illustrate the present invention and are not intended to limit the scope of protection of the present invention. Within the scope of the technical solution described in the present invention, various simple deformations and modifications made by technicians in the relevant technical field should be included in the scope of the above patent application.

Claims

1. A non-line-of-sight signal recognition method based on a channel feature weighted model, characterized in that: Contains the following steps: (1) Obtaining training samples: Collect N channel impulse response samples with labeled categories as the training set S = {(x1(t), y1), (x2(t), y2), ..., (x N (t),y N )}; where x n (t), n=1,2,…,N, represents the nth sample in the training set, which is a 1×T vector, where T represents the time dimension of the sample, y n Represents sample x n (t) label, y n ∈{0,1}; (2) Feature construction: 9 channel features are constructed from the training set samples, of which 7 basic channel features are: total energy ε, maximum amplitude r max , rise time t rise , kurtosisκ, average additional delay t med , RMS delay spread t rms and skewness ske; and additionally construct two new characteristic parameters, the attenuation factor Mc and the peak time factor IDiff, which are constructed as follows: (1) Attenuation factor Where fp max =max{fp1,fp2,fp3}, where fp1, fp2 and fp3 represent the amplitudes of the three most adjacent signals after the first path; (2) Peak time factor IDiff=|t fp -t pp | (9) Where, t fp is the path position of the first path, t pp is the path position of the peak; The attenuation factor Mc quantifies the attenuation degree of the first path amplitude. If the value is relatively large, it means that the channels are more likely to be in line-of-sight. The peak time factor IDiff measures the difference between the first path and the peak path position under different channels. In the line-of-sight case, the difference between the first path and the peak path position is much closer than that in the non-line-of-sight case. Therefore, this metric is used to judge the possibility of non-line-of-sight. Thus, a feature set F∈R consisting of 9 features is constructed. N×9 ,F={ε,r max ,t rise ,κ,t med ,t rms ,ske,Mc,IDiff}; (3) Feature preprocessing: normalize each channel feature; (IV) Establish a feature-weighted non-line-of-sight signal recognition model: Use the Relief algorithm to calculate the weights of each feature in the feature set F to form a feature weight vector w∈R 1×9 The corresponding elements in the feature set F are weighted to obtain a weighted feature set F′ = F × diag(w); the training set S is trained using a support vector machine (SVM), where the input of the SVM is the new feature set F′ weighted by the relief algorithm to solve the problem of ignoring potential key information when directly using the original feature set F; (5) Generating non-line-of-sight signal recognition results: placing the test sample into a non-line-of-sight signal recognition model based on channel feature weighting to obtain the final non-line-of-sight signal recognition results.

2. The non-line-of-sight signal recognition method based on a channel feature weighted model according to claim 1, wherein: In the step (1), N channel impulse response samples with signal category labels are collected as a training set S = {(x1(t), y1), (x2(t), y2), ..., (x N (t),y N )}; where x n (t), n = 1, 2, ..., N, represents the nth sample in the training set, which is a 1 × T vector, where T represents the time dimension of the sample; y n Represents sample x n (t) label, y n ∈{0,1}.

3. The non-line-of-sight signal recognition method based on a channel feature weighted model according to claim 2, wherein: In the step (2), features are constructed and extracted from the training set samples. First, 7 basic channel features are calculated. For a general training sample x(t), t=t1, t2, ... t T , the mathematical model of its characteristic parameters is expressed as follows: (1) Total energy (2) Maximum amplitude r max|x(t i )| max (2) (3) Rise time t rise =t H -t L (3) Among them, t H Indicates the time required for the signal amplitude to reach 0.9 times the maximum amplitude, t L Indicates the time required for the signal amplitude to reach 0.1 times the maximum amplitude; (4) Kurtosis (5) Average additional delay (6) RMS delay spread (7) Skewness Among them, x(t i ) indicates that the sample x(t) is at t=t i The value at the moment, μ |x| represents the mean of the sample x(t), σ |x| represents the standard deviation of the sample x(t).

4. The non-line-of-sight signal recognition method based on a channel feature weighted model according to claim 3, wherein: In the step (3), each feature extracted based on the channel impulse response is normalized, and the normalization formula is: In the formula, X is the original feature data, X min is the minimum value in the original feature data, X max is the maximum value in the original feature data.

5. The non-line-of-sight signal recognition method based on a channel feature weighted model according to claim 4, wherein: In the step (4), first, for the j-th feature F j ∈R N×1 , where j = 1 to 9, the weight w(j) of each feature is calculated using the Relief algorithm, and the feature weight vector w = [w(1), w(2), …, w(9)] is obtained. The calculation formula for each feature weight is as follows: In the formula, the initial weight of w(j) is 0, w k (j) is the weight of the jth feature in the feature set F in the kth update, f∈F j ,didd(F j ,f,H) and diff(F j ,f,M) respectively select the feature element f and its similar nearest neighbor sample H and different class nearest neighbor sample M in feature F j The difference between the two, m is the number of sample sampling; Furthermore, the feature weight vector w is multiplied by the corresponding element in the training set to obtain the weighted feature set F′=F×diag(w), where diag represents the operator that generates a diagonal matrix; The training set is trained using SVM, where the input of SVM is the new feature set F′ weighted by the Relief algorithm. The weight coefficient is introduced to fully explore the importance of different features to the classification model.

6. The non-line-of-sight signal recognition method based on a channel feature weighted model according to claim 5, wherein: In the step (5), the test sample is put into a non-line-of-sight signal recognition model based on channel feature weighting to obtain a final recognition result.

Images