Ultra-wideband non-line-of-sight signal identification method and device based on deep learning
By constructing a feature vector sequence and using LSTM layers and fully connected layers to process the channel impulse response signal, the problem of accurate identification of UWB positioning algorithm under non-line-of-sight conditions is solved, achieving high-precision signal discrimination and improved positioning accuracy.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- WUHAN UNIV
- Filing Date
- 2026-03-31
- Publication Date
- 2026-05-19
AI Technical Summary
Existing UWB positioning algorithms struggle to accurately identify signals under non-line-of-sight conditions, resulting in large positioning errors. Furthermore, traditional methods exhibit poor environmental adaptability, and deep learning models lack the ability to effectively capture the temporal characteristics of signals.
By constructing a feature vector sequence and inputting it into a trained non-line-of-sight signal recognition model, the channel impulse response signal is processed using LSTM layers, fully connected layers, and output layers to extract time-domain, frequency-domain, and statistical features, thereby achieving accurate identification of signal types.
It achieves accurate and robust non-line-of-sight signal discrimination in complex and ever-changing environments, improves recognition accuracy and environmental generalization ability, alleviates positioning errors, and provides a reliable solution for high-precision positioning.
Smart Images
Figure CN122069481A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of wireless communication technology, and in particular to an ultra-wideband non-line-of-sight signal recognition method and apparatus based on deep learning. Background Technology
[0002] With the booming development of the Internet of Things, intelligent navigation, and indoor positioning, the demand for high-precision location information is increasing. Ultra-wideband (UWB) technology, with its advantages such as high temporal resolution and strong multipath resistance, has become one of the key technologies for achieving high-precision positioning. In practical application scenarios, such as complex environments like large shopping malls, industrial plants, and underground parking lots, signal propagation often encounters non-line-of-sight (NLOS) situations, meaning that signals cannot propagate directly between the transmitter and receiver in a straight line, but instead reach the receiver through reflection, scattering, and other means.
[0003] Traditional UWB positioning algorithms are mostly based on the Line-of-Sight (LOS) propagation model assumption. When NLOS exists, the signal propagation delay, amplitude, and other characteristics change significantly, leading to large errors in LOS-based positioning algorithms, and even positioning failure. Currently, some related technologies can be used to identify non-line-of-sight signals, but these existing technologies have many shortcomings, including: some methods rely on simple signal feature thresholds for judgment, which are difficult to accurately determine in different environments, resulting in poor environmental adaptability; some early deep learning-based discrimination models, such as ordinary neural networks, lack the ability to effectively capture the temporal features of signals and cannot fully utilize the dynamic changes of UWB signals during multipath propagation, making it difficult to cope with complex and ever-changing real-world scenarios, and thus failing to meet the requirements for accurate NLOS discrimination under high-precision positioning needs.
[0004] Therefore, there is currently a lack of a method that is highly adaptable to the environment and can accurately identify ultra-wideband non-line-of-sight signals. Summary of the Invention
[0005] This application provides a method and apparatus for ultra-wideband non-line-of-sight signal recognition based on deep learning, to overcome the shortcomings of the aforementioned related technologies. The technical solution is as follows: In a first aspect, this application provides a deep learning-based method for ultra-wideband non-line-of-sight signal recognition, comprising: Channel impulse response data is acquired using an ultra-wideband receiver. The channel impulse response data includes channel impulse response signals at multiple consecutive moments. The channel impulse response signal at each time step is preprocessed to obtain the preprocessed channel impulse response signal. Feature extraction is performed on the preprocessed channel impulse response signal. Feature vectors for corresponding time moments are constructed based on the extracted features. A sequence of feature vectors is then constructed based on the feature vectors for each time moment. The feature vector sequence is input into the trained non-line-of-sight signal recognition model to obtain the signal type output by the non-line-of-sight signal recognition model.
[0006] In one alternative embodiment of the first aspect, the preprocessing of the channel impulse response signal at each time moment to obtain a preprocessed channel impulse response signal includes: The channel impulse response signal at each time step is low-pass filtered to remove high-frequency noise and power frequency interference, resulting in the filtered channel impulse response signal. The filtered channel impulse response signal is normalized to map the signal amplitude to the [0, 1] interval, thus obtaining the normalized channel impulse response signal. The normalized channel impulse response signal is the preprocessed channel impulse response signal.
[0007] In one alternative embodiment of the first aspect, the feature extraction based on the preprocessed channel impulse response signal, and the construction of a feature vector corresponding to the time based on the extracted features, includes: The signal amplitude of the preprocessed channel impulse response signal at each sampling point is extracted to obtain the time-domain characteristics of the signal. The time-domain characteristics include the peak value of the signal amplitude, the mean value of the signal amplitude, the variance of the signal amplitude, and the root mean square value of the signal amplitude. Based on the preprocessed channel impulse response signal, a discrete Fourier transform is performed to obtain the frequency domain characteristics of the signal, including the signal spectrum peak characteristics, the signal spectrum centroid characteristics, and the signal spectrum bandwidth characteristics. Based on the preprocessed channel impulse response signal at each sampling point, the signal amplitude, the mean of the signal amplitude, and the variance of the signal amplitude are used to calculate the statistical characteristics of the signal, including signal skewness and signal kurtosis. The feature vectors at corresponding times are constructed by arranging them in the order of time-domain features, frequency-domain features, and statistical features.
[0008] In one alternative embodiment of the first aspect, the non-line-of-sight signal recognition model includes an LSTM layer, a fully connected layer, and an output layer; The step of inputting the feature vector sequence into the trained non-line-of-sight signal recognition model and obtaining the signal type output by the non-line-of-sight signal recognition model includes: The LSTM layer receives the feature vector at time t in the feature vector sequence and the hidden state vector at time t-1 obtained by the LSTM layer, processes them to obtain the hidden state vector at time t, and outputs the hidden state vector at the last time. The hidden state vector at the last time step is input into the fully connected layer for linear transformation and nonlinear activation, and mapped to obtain the feature representation vector for classification. The feature representation vector is input into the output layer, and the channel impulse response data is classified according to the output layer to output the signal type; When t=1, the LSTM layer processes the feature vector at time 1 and the preset initial hidden state vector to obtain the hidden state vector at time 1.
[0009] In one alternative embodiment of the first aspect, the step of inputting the feature representation vector into the output layer, classifying the channel impulse response data according to the output layer, and outputting signal types includes: The output layer transforms the input feature representation vector into a probability distribution for each signal type and outputs the probability of each signal type. The signal type corresponding to the maximum probability is taken as the classification result, and the classified signal type is output.
[0010] In one alternative embodiment of the first aspect, the training process of the non-line-of-sight signal recognition model includes: A sample set is constructed by acquiring channel impulse response data containing real signal types; wherein, the sample input of each sample in the sample set is a feature vector sequence constructed based on the channel impulse response data, and the sample label is the corresponding real signal type; The non-line-of-sight signal recognition model is trained based on the sample set to obtain the prediction results obtained by the non-line-of-sight signal recognition model based on the sample input. Construct a loss function based on the difference between the prediction result and the sample label for each sample input; The model parameters of the non-line-of-sight signal recognition model are iteratively updated based on the loss function until the model converges, resulting in a well-trained non-line-of-sight signal recognition model.
[0011] In one alternative embodiment of the first aspect, the loss function is expressed as follows: ; Where L represents the loss function, Q represents the total number of samples, and q represents the sample index. This represents the sample label of the q-th sample. This represents the probability that the sample input predicted by the non-line-of-sight signal recognition model based on the q-th sample is a non-line-of-sight signal.
[0012] Secondly, this application also provides a deep learning-based ultra-wideband non-line-of-sight signal recognition device, comprising: The data acquisition module is used to acquire channel impulse response data through an ultra-wideband receiver. The channel impulse response data includes channel impulse response signals at multiple consecutive moments. The data processing module is used to preprocess the channel impulse response signal at each time moment to obtain the preprocessed channel impulse response signal. The data processing module is also used to extract features based on the preprocessed channel impulse response signal, construct feature vectors for corresponding time moments based on the extracted features, and construct a feature vector sequence based on the feature vectors for each time moment. The data recognition module is used to input the feature vector sequence into the trained non-line-of-sight signal recognition model and obtain the signal type output by the non-line-of-sight signal recognition model.
[0013] Thirdly, this application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method provided by the first aspect of this application or any implementation thereof.
[0014] Fourthly, this application also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method provided by the first aspect of this application or any implementation thereof.
[0015] The beneficial effects of the technical solution provided in this application include at least the following: This application achieves effective capture of the temporal characteristics of ultra-wideband channel impulse response signals by constructing a feature vector sequence and inputting it into a trained non-line-of-sight signal recognition model. It overcomes the shortcomings of traditional methods that rely on a single threshold, resulting in poor environmental adaptability, and the difficulty of ordinary neural networks in utilizing dynamic signal change information.
[0016] Specifically, this application preprocesses and extracts features from the channel impulse response signals at multiple consecutive moments, transforming the evolution of the signal in the time dimension into a sequence of feature vectors. This enables the deep learning model to fully learn the temporal dependencies of multipath structures under non-line-of-sight propagation conditions, thereby achieving accurate and robust non-line-of-sight signal discrimination in complex and ever-changing environments (such as large shopping malls, industrial plants, underground parking lots, etc.).
[0017] Compared with existing technologies, this application does not require a preset fixed threshold and can adapt to the signal feature distribution in different scenarios, significantly improving the accuracy and environmental generalization ability of non-line-of-sight recognition. It provides reliable prior information on signal types for subsequent positioning algorithms based on line-of-sight models, effectively alleviates positioning errors caused by non-line-of-sight propagation, and ultimately provides a reliable solution to the non-line-of-sight discrimination problem in the field of high-precision positioning. Attached Figure Description
[0018] To more clearly illustrate the technical solutions in this application or related technologies, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0019] Figure 1 This is a flowchart illustrating a deep learning-based ultrawideband non-line-of-sight signal recognition method provided in an embodiment of this application. Figure 2 This is a schematic diagram of the loss function curve of an ultra-wideband non-line-of-sight signal recognition method based on deep learning provided in an embodiment of this application; Figure 3 This is an error diagram illustrating an ultra-wideband non-line-of-sight signal recognition method based on deep learning provided in an embodiment of this application; Figure 4 This is a schematic diagram of the structure of an ultra-wideband non-line-of-sight signal recognition device based on deep learning, provided in an embodiment of this application. Figure 5 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation
[0020] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0021] The terms "comprising" and "having," and any variations thereof, in the specification, claims, and accompanying drawings of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or modules is not limited to the steps or modules listed, but may optionally include steps or modules not listed, or may optionally include other steps or modules inherent to such process, method, product, or apparatus.
[0022] It should be noted that the terms "first" and "second" used in this application are merely to distinguish similar objects and do not represent a specific ordering of the objects. It is understood that "first" and "second" can be interchanged in a specific order or sequence where permitted. It should be understood that the objects distinguished by "first" and "second" can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in an order other than those described or illustrated herein.
[0023] The present application will now be described in detail with reference to specific embodiments.
[0024] Next, combine Figure 1 This paper introduces a deep learning-based ultrawideband non-line-of-sight signal recognition method provided by embodiments of this application. For details, please refer to... Figure 1 , Figure 1 This illustration shows a flowchart of a deep learning-based ultra-wideband non-line-of-sight signal recognition method provided in an embodiment of this application. Figure 1 As shown, the method includes the following steps: S101, the channel impulse response data is acquired by an ultra-wideband receiver. The channel impulse response data includes channel impulse response signals at multiple consecutive moments. S102, preprocess the channel impulse response signal at each time step to obtain the preprocessed channel impulse response signal; S103, feature extraction is performed based on the preprocessed channel impulse response signal, feature vectors are constructed based on the extracted features, and a feature vector sequence is constructed based on the feature vectors at each time. S104, input the feature vector sequence into the trained non-line-of-sight signal recognition model to obtain the signal type output by the non-line-of-sight signal recognition model.
[0025] In some embodiments, in S101, multiple UWB transceiver nodes can be deployed in the target application scenario (such as multi-story office buildings, warehouses, etc.). Each UWB transceiver node can synchronously collect channel impulse response (CIR) signals according to a preset time interval. For example, the channel impulse response signals of multiple sampling points can be represented as a CIR signal sequence h(n), n=1, 2, ..., N, where N is the number of CIR data sampling points.
[0026] Next, in S102, the channel impulse response signal at each time step can be low-pass filtered to remove high-frequency noise and power frequency interference, thus obtaining the filtered channel impulse response signal.
[0027] For example, the channel impulse response signal can be low-pass filtered using an M-order moving average filter. The formula for calculating the filtered channel impulse response signal is as follows: ; in, The filter window length is selected based on the actual noise frequency characteristics.
[0028] Furthermore, based on the filtered channel impulse response signal, normalization processing is performed to map the signal amplitude to the [0, 1] interval, making the data collected from different batches and different devices comparable. The formula for the normalized channel impulse response signal can be expressed as:
[0029] in, , represents the maximum amplitude of the filtered channel impulse response signal; , representing the minimum amplitude of the filtered channel impulse response signal.
[0030] Therefore, the normalized channel impulse response signal at each sampling point can be obtained, expressed as: n = 1, 2, ..., N.
[0031] Based on the normalized channel impulse response signal, step S103 is performed to extract features, including: S1031, extract the signal amplitude of the preprocessed channel impulse response signal at each sampling point to obtain the time-domain characteristics of the signal. The time-domain characteristics include the peak value of the signal amplitude, the mean value of the signal amplitude, the variance of the signal amplitude, and the root mean square of the signal amplitude.
[0032] Among them, the peak characteristics of the signal amplitude are calculated. The formula is expressed as: ; Calculate the mean characteristics of the signal amplitude The formula is expressed as: ; Calculate the variance of the signal amplitude The formula is expressed as: ; Calculate the root mean square of the signal amplitude The formula is expressed as: ; Time-domain features may also include rise time. and fall edge time The rise time is the time it takes for the signal to rise from 10% of the peak value to 90% of the peak value, and the fall time is the time it takes for the signal to fall from 10% of the peak value to 90% of the peak value.
[0033] S1032, Perform a discrete Fourier transform on the preprocessed channel impulse response signal to obtain the frequency domain representation of the signal, expressed by the formula: ; Furthermore, frequency domain features of the signal are extracted based on the frequency domain representation. These frequency domain features include signal spectrum peak features, signal spectrum centroid features, and signal spectrum bandwidth features.
[0034] Among them, the peak characteristics of the signal spectrum are calculated. The formula is expressed as: ; Calculate the centroid characteristics of the signal spectrum The formula is expressed as: ; Calculate the spectral bandwidth characteristics of frequency signals The formula is expressed as: ; S1033, based on the signal amplitude, mean signal amplitude, and variance of the preprocessed channel impulse response signal at each sampling point, the statistical characteristics of the signal are calculated, including signal skewness and signal kurtosis. The formula for calculating signal skewness is as follows: ; The formula for calculating signal kurtosis is expressed as: ; S1034, arranged in the order of time-domain features, frequency-domain features, and statistical features, constructs the feature vector for the corresponding time point, expressed as the following formula: ; Here, the subscript t represents the feature vector corresponding to the channel impulse response signal acquired at time t. Based on the channel impulse response signal acquired at each time, the processing steps S1031-S1034 are performed respectively to obtain the feature vector at each time. After concatenation, the feature vector sequence can be constructed. , where T is the number of samplings of the channel impulse response signal within the time window. This feature vector sequence preserves the temporal variation information of the CIR signal, which can provide a basis for the subsequent LSTM model to learn the temporal dependencies.
[0035] Further, step S104 is executed, which processes the input feature vector sequence using a non-line-of-sight signal recognition model.
[0036] In some embodiments, the non-line-of-sight signal recognition model includes a Long Short-Term Memory (LSTM) network layer, a fully connected layer, and an output layer connected in sequence; The processing steps of the non-line-of-sight signal recognition model include the following: S1041, the LSTM layer takes the feature vector sequence as input. At each time step t, the LSTM layer processes the feature vector at time t in the feature vector sequence through the LSTM layer gating mechanism. The hidden state vector at time t-1 obtained by processing with the LSTM layer The hidden state vector at time t is obtained through processing. ; The hidden state vector at the last time step is obtained by processing through an LSTM layer.
[0037] During the input processing at each time step, the calculation formula is expressed as follows: Forgotten Gate: ; Input Gate: ; Candidate memories: ; Memory update: ; Output gate: ; Hidden state: ; in, is the Sigmoid activation function, tanh() is the hyperbolic tangent activation function, and ⊙ represents element-wise multiplication (Hadamard product). , , , These are the weight matrices for each gating mechanism. , , , For the corresponding bias vector, Let be the state of the memory cell at time t. The output is the hidden state vector at time t.
[0038] It should be noted that when t=1, the LSTM layer processes the feature vector at time 1 and the preset initial hidden state vector to obtain the hidden state vector at time 1.
[0039] S1042, the hidden state vector at the last time step is input into the fully connected layer for linear transformation and nonlinear activation, mapping to obtain a feature representation vector for classification; specifically including: ; in, This is the weight matrix of the fully connected layer. Here is the bias vector, and ReLU() is the modified linear unit activation function. This is the output vector of the fully connected layer.
[0040] S1043 inputs the feature representation vector into the output layer, classifies the channel impulse response data according to the output layer, and outputs the signal type, specifically including: The output layer transforms the input feature representation vector into a probability distribution for each signal type and outputs the probability for each signal type.
[0041] For example, in embodiments of this application used to identify whether a signal belongs to a line-of-sight (LOS) signal or a non-LOS signal, the output layer can be used to convert it into a binary classification problem of LOS signal (category y=0) and non-LOS signal (category y=1). The formula for the output probability is expressed as: , ; in, These are the components of the output vector z corresponding to category c. This indicates that the exponent value of the component in the output vector z corresponding to category c is calculated; the subscript y indicates the category, y=1 corresponds to non-line-of-sight signal, and y=0 corresponds to line-of-sight signal. This represents the summation of the exponential values of the two types of output vectors.
[0042] Then, select the signal type corresponding to the maximum probability. To output the classification result, the signal type obtained from the classification is shown in the following formula: ; For example, when the signal type corresponding to the maximum probability. When the signal is a line-of-sight signal, the output classification result is a line-of-sight signal; otherwise, the output classification result is a non-line-of-sight signal.
[0043] In some embodiments, the training process of the non-line-of-sight signal recognition model includes: S201, acquire channel impulse response data containing real signal types, and construct a sample set; wherein, the sample input of each sample in the sample set is a feature vector sequence constructed based on the channel impulse response data, and the sample label is the corresponding real signal type.
[0044] Specifically, when constructing the sample set, a large amount of CIR data can be collected in actual application scenarios using a UWB receiver. This data covers signal samples in different locations and under different obstruction conditions, thereby ensuring sample diversity.
[0045] Optionally, the sample set can be divided into a training set, a validation set, and a test set in proportions of 70%, 15%, and 15%, respectively.
[0046] S202, the non-line-of-sight signal recognition model is trained based on the sample set to obtain the prediction result of the non-line-of-sight signal recognition model based on the sample input.
[0047] S203, construct a loss function based on the difference between the prediction result and the sample label corresponding to each sample input.
[0048] S204. The model parameters of the non-line-of-sight signal recognition model are iteratively updated according to the loss function until the model converges, and the trained non-line-of-sight signal recognition model is obtained.
[0049] In some embodiments, steps S202-S203 can be performed based on the training set, with binary cross-entropy as the loss function. In S204, the Adam optimization algorithm can be used to backpropagate and update the model parameters.
[0050] The formula for the loss function is expressed as follows: ; Where L represents the loss function, Q represents the total number of samples, and q represents the sample index. This represents the sample label of the q-th sample. Let L represent the probability that the input sample predicted by the non-line-of-sight signal recognition model based on the q-th sample is a non-line-of-sight signal. The training objective is to minimize this loss function L.
[0051] The model can be trained for a certain number of rounds, such as 10 rounds, and its accuracy, recall, and other metrics can be evaluated on the validation set. The model's hyperparameters can be dynamically adjusted based on the validation results. If signs of overfitting are found, a Dropout regularization layer can be introduced or the number of LSTM layers can be adjusted.
[0052] The rules for adjusting model hyperparameters can be expressed as the following formula: ; ; ; ; ; in, For the current gradient, and These are the exponential moving average estimates of the first and second moments, respectively. , The attenuation coefficient (can be taken as...) =0.9, =0.999). For learning rate, The numerical stability constant (can be taken as...) ), These are the model parameters.
[0053] After training, the model performance can be finally evaluated using the test set, calculating comprehensive metrics such as accuracy, precision, recall, and F1 score to ensure the model achieves the expected discrimination accuracy. Figure 2 The figure shows the changes in the loss function curves of the training set and the validation set during the model training process.
[0054] It should be noted that, as an application example of this invention, the discrimination result can be used as a reference by UWB positioning systems, which can adaptively adjust their positioning algorithm strategies based on the discrimination result. For example, when the discrimination result is NLOS, the positioning system can introduce additional correction factors or switch to an algorithm mode more adapted to NLOS to reduce the negative impact of non-line-of-sight propagation on the positioning result. The specific adjustment strategies of the positioning algorithm described above are downstream applications of the discrimination result of this invention and do not constitute the core content of the method of this invention.
[0055] In some specific embodiments, simulation experiments of the method provided by this invention were conducted in office buildings, warehouses, and underground parking lots, respectively, and the average positioning error in the three types of real-world scenarios was calculated. Please refer to [the relevant documentation]. Figure 3 , Figure 3 An error diagram illustrating the method provided in the embodiments of this application is shown. By using the method of the present invention as support for the positioning algorithm, the accuracy of the positioning system can be effectively improved compared with the prior art.
[0056] The following are apparatus embodiments of this application, which can be used to execute the method embodiments of this application. For details not disclosed in the apparatus embodiments of this application, please refer to the method embodiments of this application.
[0057] Please see below. Figure 4 The image below is a schematic diagram of an ultra-wideband non-line-of-sight signal recognition device based on deep learning, provided as an exemplary embodiment of this application. The device includes: The data acquisition module is used to acquire channel impulse response data through an ultra-wideband receiver. The channel impulse response data includes channel impulse response signals at multiple consecutive moments. The data processing module is used to preprocess the channel impulse response signal at each time moment to obtain the preprocessed channel impulse response signal. The data processing module is also used to extract features based on the preprocessed channel impulse response signal, construct feature vectors for corresponding time moments based on the extracted features, and construct a feature vector sequence based on the feature vectors for each time moment. The data recognition module is used to input the feature vector sequence into the trained non-line-of-sight signal recognition model and obtain the signal type output by the non-line-of-sight signal recognition model.
[0058] It should be noted that the apparatus provided in the above embodiments, when executing the deep learning-based ultra-wideband non-line-of-sight signal recognition method, is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. Furthermore, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and their implementation process is detailed in the method embodiments, which will not be repeated here.
[0059] This application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of any of the methods described above.
[0060] Please see Figure 5 This is a structural block diagram of an electronic device provided in an embodiment of this application.
[0061] like Figure 5 As shown, the electronic device includes a processor and a memory.
[0062] In this embodiment, the processor is the control center of the computer system, and can be a processor of a physical machine or a processor of a virtual machine. The processor may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor can be implemented using at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), or PLA (Programmable Logic Array).
[0063] A processor can also include a main processor and a coprocessor. The main processor is used to process data in the wake-up state and is also called the CPU (Central Processing Unit). The coprocessor is a low-power processor used to process data in the standby state.
[0064] The memory may include one or more computer-readable storage media, which may be non-transitory. The memory may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices. In some embodiments of this application, the non-transitory computer-readable storage media in the memory are used to store at least one instruction, which is executed by a processor to implement the methods in the embodiments of this application.
[0065] In some embodiments, the electronic device further includes a peripheral device interface and at least one peripheral device. The processor, memory, and peripheral device interface are connected via a bus or signal line. Each peripheral device is connected to the peripheral device interface via a bus, signal line, or circuit board. Specifically, the peripheral device includes: a display screen, a camera, and audio circuitry. The peripheral device interface can be used to connect at least one I / O (Input / Output) related peripheral device to the processor and memory.
[0066] In some embodiments of this application, the processor, memory, and peripheral device interfaces are integrated on the same chip or circuit board; in other embodiments of this application, any one or two of the processor, memory, and peripheral device interfaces can be implemented on separate chips or circuit boards. This application does not specifically limit the implementation in this regard.
[0067] The electronic device structural block diagrams shown in the embodiments of this application do not constitute a limitation on the electronic device. The electronic device may include more or fewer components than shown, or combine certain components, or use different component arrangements.
[0068] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the methods in any of the foregoing embodiments. The computer-readable storage medium may include, but is not limited to, any type of disk, including floppy disks, optical disks, DVDs, CD-ROMs, microdrives, as well as magneto-optical disks, ROMs, RAMs, EPROMs, EEPROMs, DRAMs, VRAMs, flash memory devices, magnetic cards or optical cards, nanosystems (including molecular memory ICs), or any type of medium or device suitable for storing instructions and / or data.
[0069] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of software products. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.
[0070] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.
Claims
1. A deep learning-based method for ultra-wideband non-line-of-sight signal recognition, characterized in that, include: Channel impulse response data is acquired using an ultra-wideband receiver. The channel impulse response data includes channel impulse response signals at multiple consecutive moments. The channel impulse response signal at each time step is preprocessed to obtain the preprocessed channel impulse response signal. Feature extraction is performed on the preprocessed channel impulse response signal. Feature vectors for corresponding time moments are constructed based on the extracted features. A sequence of feature vectors is then constructed based on the feature vectors for each time moment. The feature vector sequence is input into the trained non-line-of-sight signal recognition model to obtain the signal type output by the non-line-of-sight signal recognition model.
2. The ultra-wideband non-line-of-sight signal recognition method based on deep learning according to claim 1, characterized in that, The preprocessing of the channel impulse response signal at each time moment to obtain the preprocessed channel impulse response signal includes: The channel impulse response signal at each time step is low-pass filtered to remove high-frequency noise and power frequency interference, resulting in the filtered channel impulse response signal. The filtered channel impulse response signal is normalized to map the signal amplitude to the [0, 1] interval, thus obtaining the normalized channel impulse response signal. The normalized channel impulse response signal is the preprocessed channel impulse response signal.
3. A deep learning-based ultra-wideband non-line-of-sight signal recognition method according to claim 1 or 2, characterized in that, The feature extraction based on the preprocessed channel impulse response signal, and the construction of a feature vector corresponding to the time step based on the extracted features, includes: The signal amplitude of the preprocessed channel impulse response signal at each sampling point is extracted to obtain the time-domain characteristics of the signal. The time-domain characteristics include the peak value of the signal amplitude, the mean value of the signal amplitude, the variance of the signal amplitude, and the root mean square value of the signal amplitude. Based on the preprocessed channel impulse response signal, a discrete Fourier transform is performed to obtain the frequency domain characteristics of the signal, including the signal spectrum peak characteristics, the signal spectrum centroid characteristics, and the signal spectrum bandwidth characteristics. Based on the preprocessed channel impulse response signal at each sampling point, the signal amplitude, the mean of the signal amplitude, and the variance of the signal amplitude are used to calculate the statistical characteristics of the signal, including signal skewness and signal kurtosis. The feature vectors at corresponding times are constructed by arranging them in the order of time-domain features, frequency-domain features, and statistical features.
4. The ultra-wideband non-line-of-sight signal recognition method based on deep learning according to claim 3, characterized in that, The non-line-of-sight signal recognition model includes an LSTM layer, a fully connected layer, and an output layer; The step of inputting the feature vector sequence into the trained non-line-of-sight signal recognition model and obtaining the signal type output by the non-line-of-sight signal recognition model includes: The LSTM layer receives the feature vector at time t in the feature vector sequence and the hidden state vector at time t-1 obtained by the LSTM layer, processes them to obtain the hidden state vector at time t, and outputs the hidden state vector at the last time. The hidden state vector at the last time step is input into the fully connected layer for linear transformation and nonlinear activation, and mapped to obtain the feature representation vector for classification. The feature representation vector is input into the output layer, and the channel impulse response data is classified according to the output layer to output the signal type; When t=1, the LSTM layer processes the feature vector at time 1 and the preset initial hidden state vector to obtain the hidden state vector at time 1.
5. The ultra-wideband non-line-of-sight signal recognition method based on deep learning according to claim 4, characterized in that, The feature representation vector is input into the output layer, and the channel impulse response data is classified according to the output layer to output signal types, including: The output layer transforms the input feature representation vector into a probability distribution for each signal type and outputs the probability of each signal type. The signal type corresponding to the maximum probability is taken as the classification result, and the classified signal type is output.
6. The ultra-wideband non-line-of-sight signal recognition method based on deep learning according to claim 1, characterized in that, The training process of the non-line-of-sight signal recognition model includes: A sample set is constructed by acquiring channel impulse response data containing real signal types; wherein, the sample input of each sample in the sample set is a feature vector sequence constructed based on the channel impulse response data, and the sample label is the corresponding real signal type; The non-line-of-sight signal recognition model is trained based on the sample set to obtain the prediction results obtained by the non-line-of-sight signal recognition model based on the sample input. Construct a loss function based on the difference between the prediction result and the sample label for each sample input; The model parameters of the non-line-of-sight signal recognition model are iteratively updated based on the loss function until the model converges, resulting in a well-trained non-line-of-sight signal recognition model.
7. The ultra-wideband non-line-of-sight signal recognition method based on deep learning according to claim 1, characterized in that, The formula for the loss function is expressed as follows: ； Where L represents the loss function, Q represents the total number of samples, and q represents the sample index. This represents the sample label of the q-th sample. This represents the probability that the sample input predicted by the non-line-of-sight signal recognition model based on the q-th sample is a non-line-of-sight signal.
8. A deep learning-based ultra-wideband non-line-of-sight signal recognition device, characterized in that, include: The data acquisition module is used to acquire channel impulse response data through an ultra-wideband receiver. The channel impulse response data includes channel impulse response signals at multiple consecutive moments. The data processing module is used to preprocess the channel impulse response signal at each time moment to obtain the preprocessed channel impulse response signal. The data processing module is also used to extract features based on the preprocessed channel impulse response signal, construct feature vectors for corresponding time moments based on the extracted features, and construct a feature vector sequence based on the feature vectors for each time moment. The data recognition module is used to input the feature vector sequence into the trained non-line-of-sight signal recognition model and obtain the signal type output by the non-line-of-sight signal recognition model.
9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method as described in any one of claims 1 to 7.
10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 7.