Method, device and system for identifying ultra-wideband non-line-of-sight, and storage medium

Abstract

The invention discloses a method, a device and a system for identifying ultra-wideband non-line-of-sight, and a storage medium. The method comprises the following steps: S1, acquiring ultra-wideband CIR data; s2, constructing an adaptive feature fusion network according to the CIR data; wherein statistical features and high-dimensional features extracted by a convolutional neural network are fused to form an adaptive feature fusion network; and S3, inputting ultra-wideband CIR data in an actual application scene into the adaptive feature fusion network for ultra-wideband non-line-of-sight identification. By adopting the technical scheme of the invention, the numerical statistical attribute and the high-dimensional feature are fused, and effective identification of the NLOS is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of signal processing, and in particular relates to an ultra-wideband non-line-of-sight identification method and device, system, and storage medium. Background Art

[0002] Ultra-Wide Band (UWB) has a wide range of application scenarios, including indoor communications, vehicle communications, indoor and outdoor positioning, industrial environments, and medical applications. This is due to its advantages such as low power consumption, high security, anti-multipath interference, and high data transmission rate. However, the environmental complexity of UWB application scenarios results in significant interference during signal transmission, which seriously affects the reliability, accuracy, and performance of the received signal. In an ideal channel environment, the positioning error can reach the centimeter level. However, when the environment is more complex or there are obstacles such as walls between devices, the pulse signal will be refracted, scattered, or diffracted, resulting in a delay in the signal arrival time, which ultimately greatly reduces the measurement accuracy. Due to the large difference in measurement errors: the error exceeds 100 cm under non-line-of-sight (NLOS) conditions, it is very necessary to identify the propagation state before measurement and deploy appropriate models for position estimation based on the identified state. The identification of NLOS is the first step in high-precision indoor positioning and a key prerequisite for the system to adapt to different LOS obstacles in different physical environments. Therefore, there is currently no way to effectively identify NLOS. Summary of the invention

[0003] The technical problem to be solved by the present invention is to provide a method and device, system and storage medium for identifying ultra-wideband non-line-of-sight, which integrates numerical statistical attributes with high-dimensional features to achieve effective identification of NLOS.

[0004] To achieve the above object, the present invention adopts the following technical solution:

[0005] A method for identifying ultra-wideband non-line-of-sight, comprising:

[0006] Step S1, obtaining CIR data of ultra-wideband;

[0007] Step S2: construct an adaptive feature fusion network according to the CIR data; wherein the adaptive feature fusion network is formed by fusing the statistical features with the high-dimensional features extracted by the convolutional neural network;

[0008] Step S3: inputting the CIR data of the ultra-wideband in the actual application scenario into the adaptive feature fusion network for ultra-wideband non-line-of-sight recognition.

[0009] Preferably, step S2 comprises:

[0010] Use a one-dimensional convolutional neural network to extract short sequence related features from long-term CIR data;

[0011] The short sequence correlation features are combined with the same number of statistical feature nodes to form an adaptive feature fusion network.

[0012] Preferably, the statistical features of the CIR data include: total energy, strongest signal amplitude, strongest signal reception timestamp, standard deviation, signal-to-noise ratio, amplitude of the first three arriving signals, sum of the first three arriving signals, and maximum noise value.

[0013] The present invention also provides an ultra-wideband non-line-of-sight device for identifying the same, comprising:

[0014] An acquisition module, used for acquiring CIR data of ultra-wideband;

[0015] A construction module is used to construct an adaptive feature fusion network according to CIR data; wherein the adaptive feature fusion network is formed by fusing statistical features with high-dimensional features extracted by a convolutional neural network;

[0016] The recognition module is used to input the CIR data of ultra-wideband in actual application scenarios into the adaptive feature fusion network for ultra-wideband non-line-of-sight recognition.

[0017] Preferably, the building blocks include:

[0018] A first processing module is used to extract short sequence related features from long-term CIR data using a one-dimensional convolutional neural network;

[0019] The second processing module is used to combine the short sequence related features with the same number of statistical feature nodes to form an adaptive feature fusion network.

[0020] Preferably, the statistical features of the CIR data include: total energy, strongest signal amplitude, strongest signal reception timestamp, standard deviation, signal-to-noise ratio, amplitude of the first three arriving signals, sum of the first three arriving signals, and maximum noise value.

[0021] An embodiment of the present invention further provides an ultra-wideband non-line-of-sight identification system, comprising: a memory and a processor, wherein the memory stores a computer program executed by the processor, and the computer program executes an ultra-wideband non-line-of-sight identification method when executed by the processor.

[0022] An embodiment of the present invention further provides a storage medium, on which a computer program is stored, and the computer program executes the method for identifying ultra-wideband non-line-of-sight when running.

[0023] The present invention realizes effective recognition of NLOS by fusing statistical features with high-dimensional features extracted by a convolutional neural network (CNN). BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0025] Figure 1 A flow chart of a method for identifying ultra-wideband non-line-of-sight according to an embodiment of the present invention;

[0026] Figure 2 Schematic diagram of the structure of the adaptive feature fusion network. DETAILED DESCRIPTION

[0027] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0028] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0029] Embodiment 1:

[0030] like Figure 1 As shown, an embodiment of the present invention provides a method for identifying ultra-wideband non-line-of-sight, including:

[0031] Step S1, obtaining channel impulse response (CIR) data of ultra-wideband;

[0032] Step S2: construct an adaptive feature fusion network according to the CIR data; wherein the adaptive feature fusion network is formed by fusing the statistical features with the high-dimensional features extracted by the convolutional neural network;

[0033] Step S3: inputting the CIR data of the ultra-wideband in the actual application scenario into the adaptive feature fusion network for ultra-wideband non-line-of-sight recognition.

[0034] As an implementation of an embodiment of the present invention, the statistical features of CIR data include: total energy, strongest signal amplitude, strongest signal reception timestamp, standard deviation, signal-to-noise ratio, amplitude of the first three arrival signals, sum of the first three arrival signals, and maximum value of noise. Sequence data obtained from CIR by the following formula:

[0035]

[0036] Where K is the total number of multipath components, a k and τ k Represent the amplitude and time delay of the kth path respectively. τ 1 is the earliest arrival time of the path, τ strongest is the arrival time of the strongest path.

[0037] Based on the pulse signal at the receiving end, the following ten numerical statistical features are obtained:

[0038] 1. Total Energy:

[0039]

[0040] 2. Strongest signal amplitude:

[0041] c max =max(c(t))

[0042] 3. The receiving timestamp of the strongest signal:

[0043] t max =Index(c(t)=c max )

[0044] 4. Standard Deviation (STD):

[0045]

[0046] 5. Signal-to-Noise Ratio (SNR):

[0047]

[0048] Where C represents CIR data and n is the noise vector.

[0049] The signal amplitude of the first three paths: c(t) 1st ,c(t) 2nd ,c(t) 3rd

[0050] 9. The sum of the first three arrival signals:

[0051] Amplitude(1,2,3)=c(t) 1st +c(t) 2nd+c(t) 3rd

[0052] 10. Maximum value of noise:

[0053] Noise max =max(n(t))

[0054] As an implementation of the embodiment of the present invention, step S2 includes:

[0055] Use a one-dimensional convolutional neural network to extract short sequence related features from long-term CIR data;

[0056] The short sequence correlation features are combined with the same number of statistical feature nodes to form an adaptive feature fusion network.

[0057] Further, if Figure 2 As shown, the adaptive feature fusion network includes a first model and a second model.

[0058] The input data of the first model is the acquired CIR data, and the dimension of the input data is 1*1016; the first model first performs a 1D convolution operation on the CIR data. The convolution kernel size of the first convolution layer is 3, the number of convolution kernels is 8, and the output dimension is 8*1014; in order to prevent overfitting, the dropout operation is used to prevent overfitting of the model and enhance the generalization of the model, and its coefficient is 0.6; the convolution kernel size of the second convolution layer is 3, the number of convolution kernels is 16, and the output dimension is 16*1012, and the same dropout process is performed; the output layer is dimensionalized, the two-dimensional data is stretched to one dimension, the size is 5392 elements, and the high-dimensional features are fully connected and reduced to 500 elements;

[0059] The input data of the second model is consistent with the input of the first model, and the 10 statistical features of the input data are obtained by the above 10 calculation methods. In order to balance the statistical features with the high-dimensional features extracted by the first model, the 10 statistical features are expanded to 256 and then to 500 dimensions using a fully connected layer. The second model will not destroy the balance between statistical features and high-dimensional features due to the deviation of the weight coefficient. Because different data have different statistical features, an adaptive coefficient is used to adjust the weights of statistical features in different situations.

[0060] The feature combination dimension of the first model and the second model is 1000, and then it is downsampled through two fully connected layers. The first fully connected layer downsamples from 1000 to 512, and the second fully connected layer downsamples from 512 to the number of categories 2.

[0061] Furthermore, the first model extracts high-dimensional features of CIR data by using a one-dimensional convolutional network. The kernel size of the 1D convolutional layer is 3, the stride is 1, and effective padding is used. The input signal is x(n), and the result z(n) is obtained by convolution with the convolution kernel w(n) of size 3. The convolution feature can be expressed as:

[0062]

[0063] in, represents the i-th feature of the l-th layer, represents the jth feature of the (l-1)th layer, represents the weight connected to the jth feature, represents the regularization term. The activation function σ(·) is ReLU

[33] , which is often used for nonlinear transformations due to its portability and sparsity advantages.

[0064] σ(z)=max(0,z)

[0065] The Dropout mechanism allows the model to randomly ignore certain nodes based on the Bernoulli probability distribution. Generates a binary vector consisting of 0 and 1. By reducing parameter redundancy, the adaptability of the model is enhanced. Specifically, the CIR signal sequence reflects the complexity of the environment, and the complexity of different environments varies significantly. Reducing parameter dependence can be regarded as an abstraction of the complexity of the environment, while also helping to mitigate the risk of overfitting.

[0066] Furthermore, the second model extracts high-dimensional features of numerical statistical attributes through a fully connected network. Due to the large differences in the numerical distribution of each feature, there is a significant risk of feature extraction bias. To address this issue, each feature is normalized, and then the original 10 numerical statistical attributes are expanded into 500 related feature layers through a fully connected network, which are then activated. The activation method is consistent with the convolutional network to simplify the subsequent feature fusion process. At the same time, an additional parameter is introduced in the fusion area as an adaptive coefficient to dynamically assign weights through training.

[0067] Embodiment 2:

[0068] An embodiment of the present invention further provides an ultra-wideband non-line-of-sight device for identifying the same, including:

[0069] An acquisition module, used for acquiring CIR data of ultra-wideband;

[0070] A construction module is used to construct an adaptive feature fusion network according to CIR data; wherein the adaptive feature fusion network is formed by fusing statistical features with high-dimensional features extracted by a convolutional neural network;

[0071] The recognition module is used to input the CIR data of ultra-wideband in actual application scenarios into the adaptive feature fusion network for ultra-wideband non-line-of-sight recognition.

[0072] As an implementation of an embodiment of the present invention, the building block includes:

[0073] A first processing module is used to extract short sequence related features from long-term CIR data using a one-dimensional convolutional neural network;

[0074] The second processing module is used to combine the short sequence related features with the same number of statistical feature nodes to form an adaptive feature fusion network.

[0075] As an implementation method of an embodiment of the present invention, the statistical features of the CIR data include: total energy, the amplitude of the strongest signal, the receiving timestamp of the strongest signal, the standard deviation, the signal-to-noise ratio, the amplitude of the first three arriving signals, the sum of the first three arriving signals, and the maximum value of the noise.

[0076] Embodiment 3:

[0077] An embodiment of the present invention further provides an ultra-wideband non-line-of-sight identification system, comprising: a memory and a processor, wherein the memory stores a computer program executed by the processor, and the computer program executes an ultra-wideband non-line-of-sight identification method when executed by the processor.

[0078] Embodiment 4:

[0079] An embodiment of the present invention further provides a storage medium, on which a computer program is stored, and the computer program executes the method for identifying ultra-wideband non-line-of-sight when running.

[0080] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for identifying ultra-wideband non-line-of-sight, characterized in that: include: Step S1, obtaining CIR data of ultra-wideband; Step S2: construct an adaptive feature fusion network according to the CIR data; wherein the adaptive feature fusion network is formed by fusing the statistical features with the high-dimensional features extracted by the convolutional neural network; Step S3: inputting the CIR data of the ultra-wideband in the actual application scenario into the adaptive feature fusion network for ultra-wideband non-line-of-sight recognition.

2. The method for identifying ultra-wideband non-line-of-sight according to claim 1, characterized in that: Step S2 includes: Use a one-dimensional convolutional neural network to extract short sequence related features from long-term CIR data; The short sequence correlation features are combined with the same number of statistical feature nodes to form an adaptive feature fusion network.

3. The method for identifying ultra-wideband non-line-of-sight according to claim 2, characterized in that: The statistical features of CIR data include: total energy, amplitude of the strongest signal, reception timestamp of the strongest signal, standard deviation, signal-to-noise ratio, amplitude of the first three arriving signals, sum of the first three arriving signals, and maximum noise value.

4. A device for identifying ultra-wideband non-line-of-sight, characterized in that: include: An acquisition module, used for acquiring CIR data of ultra-wideband; A construction module is used to construct an adaptive feature fusion network according to CIR data; wherein the adaptive feature fusion network is formed by fusing statistical features with high-dimensional features extracted by a convolutional neural network; The recognition module is used to input the CIR data of ultra-wideband in actual application scenarios into the adaptive feature fusion network for ultra-wideband non-line-of-sight recognition.

5. The ultra-wideband non-line-of-sight identification device according to claim 4, characterized in that: The building blocks include: A first processing module is used to extract short sequence related features from long-term CIR data using a one-dimensional convolutional neural network; The second processing module is used to combine the short sequence related features with the same number of statistical feature nodes to form an adaptive feature fusion network.

6. The ultra-wideband non-line-of-sight identification device according to claim 5, characterized in that: The statistical features of CIR data include: total energy, amplitude of the strongest signal, reception timestamp of the strongest signal, standard deviation, signal-to-noise ratio, amplitude of the first three arriving signals, sum of the first three arriving signals, and maximum noise value.

7. An ultra-wideband non-line-of-sight identification system, characterized in that: include: A memory and a processor, wherein the memory stores a computer program executed by the processor, and when the computer program is executed by the processor, the method for identifying ultra-wideband non-line-of-sight according to any one of claims 1 to 3 is executed.

8. A storage medium, characterized in that: The storage medium stores a computer program, and the computer program executes the method for identifying ultra-wideband non-line-of-sight according to any one of claims 1 to 3 when running.

Images