Non-line-of-sight environment identification method

Abstract

The invention provides a non line of sight (NLOS) environment identification method based on UWB distance measurement data, and the method mainly comprises the following steps (as shown in figure 1): 1, carrying out the statistics of the effective sample rate of UWB distance measurement data samples, and identifying the distance measurement samples with the effective sample rate smaller than a threshold value c0 to be collected in an NLOS environment; 2, further checking the ranging samples which are not identified as collected in the NLOS environment, and firstly determining histogram parameters for statistical feature analysis of the ranging samples and a probability density function type for fitting sample histogram distribution; then, establishing a cumulative distribution function of a ranging sample and a cumulative distribution function of a fitting probability density function of the cumulative distribution function by using the determined histogram parameters and probability density function types; and finally, carrying out goodness of fit test by utilizing Kolmogorov-Smirnov (KS) test so as to identify the NLOS environment. According to the method, the NLOS acquisition environment is identified according to the statistical characteristics of the ranging data, and the UWB positioning system is supported to improve the positioning precision. The method is suitable for the technical field of indoor positioning.

Description

Technical Field

[0001] The present invention belongs to the field of indoor positioning and location services, and particularly relates to a non-line-of-sight environment recognition method. Background Art

[0002] UWB (Ultra-Wideband) is a wireless communication technology, which is characterized by transmitting a large amount of data at a relatively low power level. The UWB technology uses a very wide frequency band (exceeding 500 MHz) to transmit short pulse signals, and the time intervals of these pulse signals are very short, usually only a few nanoseconds to a few hundred picoseconds. The characteristic of the wide working frequency band of UWB enables it to achieve high-precision ranging and positioning capabilities. By measuring parameters such as the time of arrival of the signal and multipath propagation, UWB can achieve ranging and positioning accuracies at the centimeter level. This makes UWB have broad application prospects in applications such as indoor positioning and location services, and cargo tracking.

[0003] In a complex indoor environment, such as walls, doors, windows, furniture, etc., the UWB signal will be blocked, resulting in non-line-of-sight (NLOS) propagation, causing signal distortion, and further leading to large errors in the distance measurement values, seriously affecting the indoor positioning accuracy. In response to this, the present invention proposes a non-line-of-sight environment recognition method to identify the data collected in the non-line-of-sight environment and improve the positioning accuracy in a complex indoor environment. Summary of the Invention

[0004] The purpose of the present invention is to solve the problem that the NLOS environment seriously affects the positioning accuracy, and provides a method that can effectively identify the NLOS environment. It uses the differences in statistical characteristics between the ranging samples collected in the line-of-sight (LOS) environment and the NLOS environment to identify the NLOS environment. The differences in the statistical characteristics are achieved by establishing a frequency histogram of the distance samples and its probability density function and testing the fitting degree between the two to achieve the purpose of NLOS recognition.

[0005] To achieve the above-mentioned invention purpose, the following technical solutions are adopted: A non-line-of-sight environment recognition method, the method includes the following steps:

[0006] Step 1: Calculate the effective sample rate of the ranging samples, and the ranging samples with the effective sample rate less than the threshold are identified as being collected in the NLOS environment.

[0007] Step 2: Further test the ranging samples that are not identified as being collected in the NLOS environment. First, determine the histogram parameters for the statistical characteristic analysis of the ranging samples and the type of probability density function for fitting the sample histogram distribution.

[0008] Step 3: Using the determined histogram parameters and probability density function types, establish the cumulative distribution function of the ranging samples and the cumulative distribution function of its fitted probability density function, and identify the NLOS acquisition environment through the goodness-of-fit test.

[0009] Preferably, in Step 1, the effective sample rate is the ratio of the number of distance samples actually collected within a certain time length at the fixed sampling rate of the device to the theoretical number of samples. The theoretical number of samples is the total amount of data samples collected by the device at the fixed sampling frequency during the acquisition duration, that is

[0010]

[0011] where F is the sampling rate of the device, in Hz, t is the acquisition duration, and m represents the theoretical number of samples.

[0012] More preferably, in Step 2, the histogram parameter is the bin width, and the probability density function type is a common function type: one of the normal distribution function, stable distribution function, T distribution function, and Logistic distribution function. To determine the bandwidth and probability density function type, ranging samples under line-of-sight conditions are collected in the same environment, histograms with different bandwidth values are generated, and each common probability density function is fitted, and then the fitting residuals are calculated. Select the bandwidth value and probability density function type with the smallest fitting residual for the histogram bandwidth and its fitted probability density function for identifying the NLOS environment.

[0013] The fitting residual is the mean-squared error (MSE) between the normalized frequency corresponding to each bin of the histogram and the probability density function value at the center value of the bin value interval, that is

[0014]

[0015] where H i represents the normalized frequency corresponding to the i-th histogram bin, and E i represents the probability density function value at the center value of the i-th histogram bin value interval. K represents the number of histogram bins,

[0016]

[0017] where Max(distance) and Min(distance) are the maximum and minimum values of the ranging samples, and binWidth is the bandwidth of the histogram bin.

[0018] The normalized frequency is the ratio of the frequency corresponding to a single histogram bin to the total frequency of all bins, and its calculation formula is as follows:

[0019]

[0020] More preferably, the cumulative distribution function of the ranging samples described in step 3 is constructed as follows:

[0021] First, data standardization is performed to convert data of different scales to a unified scale, which is convenient for comparison, analysis, and modeling. The present invention uses the Z-Score standardization method to standardize the ranging samples, converting the data into a distribution with a mean of 0 and a standard deviation of 1. The calculation formula is as follows:

[0022]

[0023] where x is the ranging sample, are the mean and standard deviation of the ranging samples, respectively.

[0024] Furthermore, the cumulative distribution function of the standardized ranging samples is constructed. First, a set of standardized ranging samples is sorted. Let the standardized ranging sample set be , where is the i-th sample value. After sorting the samples, we get , where represents the i-th sample value after sorting. For any value z, the mathematical expression of the cumulative distribution function can be expressed as:

[0025]

[0026] where, is the indicator function, which takes the value of 1 when the condition is true and 0 otherwise. In other words, represents the proportion of data points in the dataset that are less than or equal to z.

[0027] More preferably, for the cumulative distribution function of the fitted probability density function described in step 3, the forms of the cumulative distribution functions of common probability density functions are known. Taking the fitted probability density function as the standard normal distribution as an example, its functional form is

[0028]

[0029] The cumulative distribution function represents the probability that the random variable X is less than or equal to x, and it is the integral of the probability density function from to x:

[0030]

[0031] Therefore, the cumulative distribution function of the standard normal distribution can be expressed as:

[0032]

[0033] More preferably, for the goodness-of-fit test described in step 3, the Kolmogorov-Smirnov test method is used to test the cumulative distribution function of the standardized ranging samples constructed previously and the cumulative distribution function of the fitted probability density function (taking the standard normal distribution as an example) to determine whether the standardized ranging samples are from the standard normal distribution. The test steps are as follows:

[0034] First step, propose the null hypothesis H0: , that is, assume that the data samples follow the theoretical distribution of this hypothesis;

[0035] Second step, calculate the maximum difference between the two cumulative distribution functions, that is, the KS statistic, usually denoted as D,

[0036]

[0037] where, represents taking the maximum difference over all data points x, and the calculation process is as follows:

[0038] First, according to the two cumulative distribution functions and , calculate the cumulative distribution function values corresponding to the standardized ranging sample set respectively, to obtain the set

[0039] { (i = 1, 2…, n)

[0040] (i = 1, 2…, n)

[0041] Then, calculate the absolute difference between the two numerical sets,

[0042] (i = 1, 2…, n)

[0043] Finally, sort the absolute difference set and find the maximum difference, which is the KS statistic D.

[0044] Third step, determine the rejection region of H0 , that is, when , H0 is not valid. Among them, represents the critical value of the rejection region at the significance level α with the sample size of . According to different experimental scenarios, different significance levels can be set to meet the requirements of test sensitivity and error risk in different scenarios. In the present invention, the significance level of 0.01 is illustrated. For a given significance level α, when When the number of ranging samples \(n\) changes, the critical value of the rejection region decreases as \(n\) increases. The exact value can be obtained by referring to a table. When the number of samples is \( \), the critical value of the rejection region needs to be approximately calculated through an expression related to \(n\). The expression is as follows:

[0045]

[0046] After determining the critical value of the rejection region, the result is judged. When holds, \(H_0\) does not hold, that is, the sample and the theoretical distribution do not belong to the same distribution or are inconsistent, and the data sample is judged to be the data collected under NLOS; if holds, \(H_0\) holds, that is, the sample and the theoretical distribution are the same distribution or are consistent, and the data sample is judged to be the data collected under LOS.

[0047] The method of the present invention starts from the UWB original ranging sample data, analyzes the data using statistical methods, and judges whether it is collected under NLOS according to the statistical characteristics of the data itself, so as to achieve the purpose of identifying the NLOS environment. Brief Description of the Drawings

[0048] Figure 1 is a schematic flow chart of the method of the present invention

[0049] Figure 2 is a schematic diagram of the test site Detailed Embodiments

[0050] The present invention will be described in detail below in conjunction with the drawings and specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and not to limit the application scope of the present invention. Operations involving detailed parameter descriptions but not specifically described in the embodiments are well known to those skilled in the art.

[0051] Embodiment 1

[0052] The working flow chart of the NLOS environment recognition method of the present invention is as shown in Figure 1 and includes the following steps:

[0053] Step 1: Calculate the effective sample rate of the ranging samples. The ranging samples with an effective sample rate less than the threshold are identified as being collected in the NLOS environment.

[0054] Step 2: Further test the ranging samples not identified as being collected in the NLOS environment. First, determine the histogram parameters for analyzing the statistical characteristics of the ranging samples and the type of probability density function for fitting the histogram distribution of the samples.

[0055] Step 3: Using the determined histogram parameters and probability density function types, establish the cumulative distribution function of the ranging samples and the cumulative distribution function of its fitted probability density function, and identify the NLOS acquisition environment through the goodness-of-fit test.

[0056] The specific steps are as follows:

[0057] 1. Raw data acquisition

[0058] The present invention is directed to the scenario where collectors use a computer and UWB devices to perform ranging and positioning in an indoor space, and the computer records the distance data between UWB tag nodes and each UWB base station node.

[0059] 2. NLOS acquisition environment identification

[0060] (1) First, calculate the effective sample rate of the ranging samples to preliminarily identify the NLOS acquisition environment. The effective sample rate is the ratio of the number of actually collected distance samples to the theoretical sample number within a certain time length at the fixed sampling rate of the device. The theoretical sample number is the total amount of data collected by the device at the fixed sampling frequency during the acquisition duration, that is

[0061]

[0062] where F is the sampling rate of the device, in Hz, t is the acquisition duration, and m represents the theoretical sample number collected within this time length.

[0063] Based on the collected theoretical sample number, calculate the proportion of the actually collected samples to the theoretical samples, that is, the effective sample rate, and identify the ranging samples with a proportion less than the threshold as being collected in the NLOS environment. That is

[0064]

[0065] where n and m represent the actually collected samples and the theoretical sample number respectively, is the threshold of the proportion of the effective data volume.

[0066] (2) Further test the ranging samples not identified as being collected in the NLOS environment. Collect the ranging samples under the LOS condition in the same environment, generate histograms for different bandwidth values, and fit them according to common probability density functions such as the normal distribution function, stable distribution function, T distribution function, and Logistic distribution function, and then calculate the fitting residuals, as shown in the following formula. Select the bandwidth value and the type of probability density function with the smallest fitting residuals as the histogram bandwidth and its fitted probability density function for identifying the NLOS environment.

[0067]

[0068] Among them, H i represents the normalized frequency corresponding to the i-th histogram bar, and E i represents the probability density function value at the center value of the value range of the i-th histogram bar. K represents the number of histogram bars,

[0069]

[0070] where Max(distance) and Min(distance) are the maximum and minimum values of the ranging samples, and binWidth is the bandwidth of the histogram bars.

[0071] The normalized frequency is the ratio of the frequency corresponding to a single histogram bar to the total frequency of all bars, and its calculation formula is as follows:

[0072]

[0073] Next, according to the parameters determined above, construct the cumulative distribution function of the ranging samples. First, use the Z-Score standardization method to standardize the ranging samples, converting the data into a distribution with a mean of 0 and a standard deviation of 1. The calculation formula is as follows:

[0074]

[0075] where x is the ranging sample, are the mean and standard deviation of the ranging samples, respectively.

[0076] Furthermore, construct the cumulative distribution function of the standardized ranging samples. First, sort a set of standardized ranging samples. Let the standardized ranging sample set be , where is the i-th sample value. After sorting the samples, we get , where represents the i-th sample value after sorting. For any value z, the mathematical expression of the cumulative distribution function can be expressed as:

[0077]

[0078] where is the indicator function, which takes the value of 1 when the condition is true and 0 otherwise. In other words, represents the proportion of data points in the dataset that are less than or equal to z.

[0079] Taking the fitting probability density function type as the standard normal distribution as an example, the construction process of the cumulative distribution function of the standardized samples is as follows:

[0080] First, determine the cumulative distribution function of the standard normal distribution:

[0081]

[0082] Then, identify NLOS through the goodness-of-fit test, that is, use the Kolmogorov-Smirnov (KS) test method to test the cumulative distribution function of the standardized ranging samples constructed previously and the cumulative distribution function of the fitted probability density function (taking the standard normal distribution as an example) to determine whether the standardized ranging samples come from the standard normal distribution by the similarity between them. The test steps are as follows:

[0083] The first step is to propose the null hypothesis H0: , that is, assume that the data samples follow the theoretical distribution of this hypothesis;

[0084] The second step is to calculate the maximum difference between the two cumulative distribution functions, that is, the KS statistic, usually denoted as D,

[0085]

[0086] where, represents taking the maximum difference over all data points x, and the calculation process is as follows:

[0087] First, according to the two cumulative distribution functions and , calculate the cumulative distribution function values corresponding to the standardized ranging sample set respectively, to obtain the set

[0088] { (i = 1, 2…, n)

[0089] (i = 1, 2…, n)

[0090] Then, calculate the absolute differences between the two numerical sets,

[0091] (i = 1, 2…, n)

[0092] Finally, sort the set of absolute differences and find the maximum difference, which is the KS statistic D.

[0093] The third step is to determine the rejection region of H0 , that is, when , H0 does not hold. Where, Denoted as the critical value of the rejection region when the significance level is α and the sample size is n. According to different experimental scenarios, different significance levels can be set to meet the requirements of test sensitivity and error risk in different scenarios. In the present invention, the significance level of 0.01 is illustrated. For a given significance level α, when When, with the change of the ranging sample number n, the critical value of the rejection region decreases with the increase of n, and the exact value can be obtained by referring to the table. And when the sample quantity When, the critical value of the rejection region needs to be approximately calculated by an expression about n, and the expression is as follows:

[0094]

[0095] After determining the critical value of the rejection region, the result is judged. When When, H0 does not hold, that is, the sample and the theoretical distribution do not belong to the same distribution or are inconsistent, and the data sample is judged as the data collected under NLOS; if When, H0 holds, that is, the sample and the theoretical distribution are the same distribution or are consistent, and the data sample is judged as the data collected under LOS.

[0096] Recognition effect test

[0097] In order to test the NLOS environment recognition effect of the inventive method, an experiment was carried out in the corridor of an office building. As Figure 2 Shown, the test site is about 20 meters long from east to west and about 15 meters long from north to south. 20 test points were evenly selected for the experiment. The final recognition effect and the average point position error calculated before and after recognition are shown in Table 1 below. The inventive method can achieve 100% recognition of NLOS environment data. The average positioning error of the point position before and after recognition drops from 3.59 m to 0.15 m, which can effectively recognize the NLOS environment and can achieve more accurate positioning.

[0098] Table 1

[0099] Number of test points Number of effectively recognized points NLOS recognition rate Average positioning error before recognition Average positioning error after recognition 20 20 100% 3.59m 0.15m

Claims

1. A non-line-of-sight environment recognition method, characterized in that: The steps include: (1) Calculate the effective sample rate of the ranging samples. The ranging samples with an effective sample rate less than the threshold c0 are identified as being collected in a non-line of sight (NLOS) environment. (2) Further inspection of ranging samples collected in environments that are not identified as NLOS. First, the histogram parameters used for statistical feature analysis of ranging samples and the type of probability density function used to fit the distribution of sample histograms are determined; (3) Using the determined histogram parameters and probability density function type, the cumulative distribution function of the ranging samples and the cumulative distribution function of its fitted probability density function are established, and the NLOS samples are identified through a goodness of fit test.

2. The non-line-of-sight environment recognition method according to claim 1, characterized in that: The effective sample rate using ranging samples in step (1) is to identify the NLOS environment by judging whether the effective sample rate is less than a set threshold c0, as follows: The effective sample rate is the ratio of the number of distance samples actually collected by the device at a fixed sampling rate within a certain time length to the theoretical number of samples. The theoretical number of samples is the total amount of data samples collected by the device at a fixed sampling frequency within the collection time length, that is, m=F×t Where F is the device sampling rate in Hz, t is the acquisition duration, and m represents the theoretical number of samples.

3. The non-line-of-sight environment recognition method according to claim 1, characterized in that: The histogram parameters and probability density function type determined in step (2) refer to the process of collecting ranging samples under line-of-sight conditions in the same environment, generating histograms with different bandwidth values, and fitting according to each commonly used probability density function, and then calculating the fitting residual, as shown in the following formula, which is the mean square error (MSE) between the normalized frequency corresponding to each strip of the histogram and the probability density function value of the center value of the strip value interval. The bandwidth value and probability density function type with the smallest fitting residual are selected for identifying the NLOS environment: Among them, H i represents the normalized frequency corresponding to the i-th histogram band, E i represents the probability density function value of the center value of the i-th histogram band value interval, K represents the number of histogram bands, Among them, Max(distance) and Min(distance) are the maximum and minimum values ​​of the distance measurement samples, and binWidth is the bandwidth of the histogram band. The normalized frequency is the ratio of the frequency corresponding to a single histogram band to the total frequency of all bands, and its calculation formula is as follows.

4. The non-line-of-sight environment recognition method according to claim 1, characterized in that: The cumulative distribution function of the ranging samples in step (3) is constructed as follows: (1) Data standardization: converting data of different scales to a unified scale to facilitate comparison, analysis and modeling. e The standardization method standardizes the ranging samples and converts the data into a distribution with a mean of 0 and a standard deviation of 1. The calculation formula is as follows: Among them, x is the ranging sample, μ and δ are the mean and standard deviation of the ranging samples respectively; (2) Construct the cumulative distribution function of the standardized distance measurement samples. First, sort a set of standardized distance measurement samples. Let the standardized distance measurement sample set be {z1, z2, …, z n }, where z i is the i-th sample value, sort the samples to get {z (1) , z (2) , …, z (n) }, where z (i) represents the i-th sample value after sorting. For any value z, the mathematical expression of the cumulative distribution function F1(z) can be expressed as: Where I(·) is an indicator function, which takes the value of 1 when the condition is true and 0 otherwise. In other words, F1(z) represents the proportion of data points in the data set that are less than or equal to z.

5. The non-line-of-sight environment recognition method according to claim 1, characterized in that: The cumulative distribution function of the probability density function of the fitting step (3) is in a known form. We take the fitting probability density function as a standard normal distribution as an example, and its function form is: The cumulative distribution function F(x) represents the probability that the random variable X is less than or equal to x. It is the integral of the probability density function from -∞ to x: Therefore, the cumulative distribution function of the standard normal distribution can be expressed as follows.

6. The non-line-of-sight environment recognition method according to claim 1, characterized in that: The goodness of fit test in step (3) is to use the Kolmogorov-Smirnov test method to test the similarity between the cumulative distribution function F1(x) of the standardized distance measurement sample constructed in the previous order and the cumulative distribution function F2(x) of the fitted probability density function (taking the standard normal distribution as an example) to determine whether the standardized distance measurement sample comes from the standard normal distribution. The test steps are as follows: (1) Propose the null hypothesis H0: F1(x) = F2(x), that is, assume that the data sample obeys this hypothesized theoretical distribution; (2) Calculate the maximum difference between the two cumulative distribution functions, namely the KS statistic, usually denoted as D; in, It means taking the maximum difference among all data points x. The calculation process is as follows: First, according to the two cumulative distribution functions F1(x) and F2(x), the standardized ranging sample sets {z1, z2, …, z n }The corresponding cumulative distribution function value is obtained by {F1(z i )}(i=1,2...,n) {F2(z i )}(i=1,2...,n) Then, calculate the absolute difference between the two sets of values {d i =|F1(z i )-{F2(z i )|}(i=1,2...,n) Finally, sort the absolute difference set and find the maximum difference, which is the KS statistic D; (3) Determine the rejection region D ≥ D of H0 n，α , that is, when D ≥ D n，α When H0 is not established, D n，α It is expressed as the rejection region critical value of the sample size n under the significance level α. Different significance levels can be set according to different experimental scenarios to meet the requirements of different scenarios for test sensitivity and error risk. The present invention illustrates the significance level of 0.

01. For a given significance level α, when n≤40, as the number of ranging samples n changes, the rejection region critical value decreases as n increases. The specific value can be obtained by looking up the table. When the number of samples n>40, the rejection region critical value needs to be approximately calculated by an expression about n. The expression is as follows: After determining the rejection region critical value, the result is judged. When D ≥ D n，α When H0 is not true, that is, the sample and the theoretical distribution do not belong to the same distribution or are inconsistent, and the data sample is judged to be data collected under NLOS. If D <D n，α , H0 is established, that is, the sample and the theoretical distribution are the same or consistent, and the data sample is judged to be the data collected under line of sight.

Images