Beidou navigation multipath signal detection method
By training a neural network model with a supervised learning algorithm based on unsupervised learning, the problems of high computational load and sensor dependence in multipath signal detection in existing technologies are solved, achieving lightweight, real-time multipath signal recognition and improving positioning accuracy in urban environments.
Patent Information
- Application Number
- CN202511909415.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-17
- Publication Date
- 2026-02-24
AI Technical Summary
Existing machine learning methods for multipath signal detection and suppression suffer from high computational load and dependence on external sensors, and labeled datasets are expensive, making it difficult to effectively address multipath effects in complex urban environments.
An unsupervised learning method is used to remove outliers, and a neural network model is trained by combining supervised learning algorithms. By collecting signal data in different static environments, a multipath signal detection model is constructed, and multipath signal identification is performed directly at the receiving end, avoiding real-time mapping and external sensors, thus reducing system deployment costs and computational load.
It significantly reduces system deployment costs and computational load, improves the accuracy of training data labels, enhances model generalization ability, enables lightweight, real-time multipath signal recognition, and enhances positioning robustness in complex urban environments.
Smart Images
Figure CN121559554A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of wireless communication technology. Background Technology
[0002] Our daily lives and work, as well as the development of various industries, increasingly rely on geolocation and navigation systems, with the BeiDou Navigation Satellite System being particularly widely used. However, various buildings in urban environments, such as high-rise buildings and moving vehicles, often cause BeiDou satellite signals to be blocked and reflected, leading to multipath effects and forming the so-called "urban canyon effect." This error poses a serious threat to accurate and reliable location services and severely affects the accuracy of satellite navigation and positioning.
[0003] Multipath effect is a common phenomenon in wireless communication. It mainly refers to the fact that during the propagation of radio signals, in addition to the direct path, one or more indirect paths are generated due to effects such as reflection, refraction, and scattering, causing the same signal to reach the receiver through different paths. Since these paths have different lengths and propagation speeds, the same signal received may have differences in phase and amplitude. This is the multipath effect.
[0004] In satellite communications, due to the complexity of the physical environment, existing multipath detection and suppression methods cannot effectively handle complex scenarios. Therefore, machine learning algorithms can play a significant role in such situations. The principle of using machine learning to achieve multipath detection and suppression can be divided into several stages: data collection, feature engineering, training the machine learning model, predicting and suppressing multipath effects, and feedback and updating. Most of these methods rely on supervised learning-based classifiers, which are limited by the need to pre-label training samples, and the accuracy of the labeling directly affects the classifier's performance. Generally speaking, obtaining an accurate label dataset that can represent all state types is very difficult and costly. Currently, the mainstream signal type labeling method is based on 3D building models. This method is not only limited by the accuracy and timeliness of the 3D building model but also requires the accurate location of the receiver to obtain the correct classification label. Ray tracing methods based on 3D building models can further determine multipath signals. However, this method has an excessively high computational load and is affected by model accuracy and building materials. In some studies that do not use 3D building models, external sensors or even manual assistance are needed to label signal types. Summary of the Invention
[0005] The purpose of this invention is to address the problem of high load and the need for external sensors in existing machine learning applications for multipath signal detection and suppression, and to propose a method for detecting multipath signals in BeiDou navigation.
[0006] The method for detecting BeiDou navigation multipath signals according to the present invention includes the following steps:
[0007] The system receives the BeiDou navigation and positioning signal from the current target environment and uses a neural network-based multipath signal detection model to detect the received signal and identify whether the BeiDou navigation and positioning signal from the current target environment contains multipath signals.
[0008] Furthermore, the multipath signal detection model is obtained through pre-training, and the training method for the multipath signal detection model is as follows:
[0009] Raw observation data of direct signals and raw observation data of multipath signals from BeiDou satellites were collected in two different static environments.
[0010] Unsupervised learning methods were used to preprocess the raw observation data of the direct signal and the raw observation data of the multipath signal to remove outliers, thereby obtaining the direct signal dataset and the multipath training dataset, respectively.
[0011] The direct-shoot dataset and the multipath training dataset are randomly shuffled and merged to construct the final training set;
[0012] Based on the final training set, a supervised learning algorithm is used to train the neural network model to obtain a multipath signal detection model based on a neural network.
[0013] Furthermore, the two different static environments are: a rooftop location with an open and unobstructed view and a location at the bottom of a tall building.
[0014] Furthermore, the unsupervised learning method is a clustering algorithm used to remove outliers from the original observation data of the direct signal and the original observation data of the multipath signal.
[0015] Furthermore, the supervised learning algorithm is either the K-nearest neighbor algorithm or the extreme gradient boosting tree algorithm.
[0016] Furthermore, the neural network model is a feedforward neural network.
[0017] Furthermore, the observation features in the raw observation data of the direct signal and the raw observation data of the multipath signal include one or more of the following: carrier phase, pseudorange, signal-to-noise ratio, and Doppler shift.
[0018] Compared with the prior art, the present invention has the following advantages:
[0019] This invention employs unsupervised learning to remove outliers from the original observation data, reducing the presence of mixed direct and multipath signals in the multipath signal dataset. This ensures more accurate labels in both the direct and multipath satellite observation datasets, guaranteeing the subsequent signal classification rule learning. The preprocessed direct and multipath satellite observation datasets are then merged. Supervised learning is then used to learn the signal classification rules; the merged dataset is shuffled and used as the final training set. After training the multipath signal detection model, the supervised learning algorithm with the best classification performance is selected for subsequent field testing in dynamic environments. This invention addresses multipath signal detection. By collecting direct and multipath signals in two typical environments—unobstructed and obstructed by tall buildings—and combining unsupervised learning to clean the data, followed by supervised learning to train the model, the receiving end can directly call the multipath signal detection model to complete multipath identification. This eliminates the need for real-time mapping or external sensors, completely freeing the system from dependence on 3D building models, ray tracing, or additional sensors, significantly reducing system deployment costs and computational load. Simultaneously, the multipath signal detection model outputs a binary result of "whether multipath is present," which can be directly embedded into existing baseband processes, achieving plug-and-play compatibility upgrades. This improves the accuracy of training data labels, enhances model generalization ability, achieves lightweight, real-time multipath signal identification, and enhances positioning robustness in complex urban environments. Attached Figure Description
[0020] Figure 1 This is a flowchart of a method for detecting multipath signals in BeiDou navigation, as described in Specific Implementation Method 1.
[0021] Figure 2 This is a flowchart of the training method for the multipath signal detection model in Implementation Method 2.
[0022] Figure 3 This is a multipath signal detection diagram for Scenario 1 in Implementation Method 2. Detailed Implementation
[0023] Specific Implementation Method 1: Combination Figure 1 This embodiment describes a method for detecting BeiDou navigation multipath signals, which includes the following steps:
[0024] The system receives the BeiDou navigation and positioning signal from the current target environment and uses a neural network-based multipath signal detection model to detect the received signal and identify whether the BeiDou navigation and positioning signal from the current target environment contains multipath signals.
[0025] In this embodiment, the receiver directly calls the pre-trained model to complete multipath identification without the need for real-time mapping or external sensors, significantly reducing computational load and hardware costs. At the same time, the multipath signal detection model outputs a binary result of "whether it contains multipath" which can be directly embedded into the existing baseband process, achieving plug-and-play compatibility upgrades.
[0026] Specific Implementation Method Two: Combination Figures 2 to 3 This embodiment further defines the method for detecting multipath signals in BeiDou navigation described in Specific Embodiment 1. In this embodiment, the multipath signal detection model is obtained through pre-training, and the training method for the multipath signal detection model is as follows:
[0027] Raw observation data of direct signals and raw observation data of multipath signals from BeiDou satellites were collected in two different static environments.
[0028] Unsupervised learning methods were used to preprocess the raw observation data of the direct signal and the raw observation data of the multipath signal to remove outliers, thereby obtaining the direct signal dataset and the multipath training dataset, respectively.
[0029] The direct-shoot dataset and the multipath training dataset are randomly shuffled and merged to construct the final training set;
[0030] Based on the final training set, a supervised learning algorithm is used to train the neural network model to obtain a multipath signal detection model based on a neural network.
[0031] In this embodiment, the technical effectiveness of this embodiment is verified through experiments using three typical urban multipath interference scenarios selected below:
[0032] Scenario 1: Contains tall buildings. The multipath signal detection map in this scenario shows that the occurrence of multipath signals is closely related to the geographical environment. When the experimental vehicle travels southwest, because the initial intersection is next to a primary school playground, located in an open and flat area, the satellite signals received by the vehicle contain relatively few multipath signals within 0 to 110 seconds. However, when the experimental vehicle enters the intersection, the satellite signals are immediately blocked by the department store and apartment building on the roadside, causing the vehicle to detect satellite signals containing multipath signals in subsequent detection periods. Furthermore, because the tall buildings along the roadside are not continuous, there are still certain distances between adjacent buildings. Therefore, in subsequent detection epochs, a small number of satellite signals still do not show obvious multipath signals. The specific multipath signal detection results are as follows: Figure 3 As shown.
[0033] Scenario 2: Contains tall green vegetation. The environment in this scenario is beside a river. The experimental vehicle travels northwest along the riverbank. Although there are no continuous, dense tall buildings around the route, there are dense trees on both sides of the path, with tree heights much higher than the vehicle's receiving antenna. Due to the vegetation cover, such as trees and forests, signal attenuation, reflection, and scattering occur, creating multiple paths. Therefore, the dense trees also result in a large amount of multipath signal in the received satellite signal. The few periods during the entire detection epoch where no multipath signal was detected were between 70s and 95s. This is likely because the vegetation cover is less extensive at the bends along the riverbank, thus reducing the generation of multipath signals.
[0034] Scenario 3: Includes a traffic bridge structure. Scenario 3 takes place in a complex, multi-dimensional environment of traffic bridges and viaducts. As the experimental vehicle drives onto the overpass, its signal quality is poor due to the obstruction caused by the viaduct, resulting in strong multipath interference. When the vehicle briefly leaves the overpass, it remains within the gaps in the viaduct, so only a few observation epochs contain weak multipath signals. However, when the vehicle enters the lanes below the Second Ring Road viaduct, it is again obstructed by the viaduct, resulting in significant multipath interference in the received signal. In subsequent observation epochs, the vehicle leaves the viaduct, but the overpass continues to obstruct the signal, causing the vehicle to be intermittently affected by the overpass during this period. Strong and weak multipath interference alternate at the receiver.
[0035] Combining prior knowledge with the multipath signal detection results from the three real-world scenarios above, it can be demonstrated that in complex and multidimensional urban environments, buildings and urban structures, vegetation cover, and man-made structures such as bridges, elevated roads, and tunnels can cause signal reflection and diffraction, resulting in multiple paths. Therefore, in urban environments, due to the influence of complex terrain, buildings, and other environmental factors, the propagation paths of satellite signals become diversified, leading to multipath signals received by the receiver. This multipath effect may affect the accuracy and reliability of positioning and navigation systems. Furthermore, the multipath signal detection results largely conform to our well-known prior knowledge, indicating that our multipath signal detection method is convincing and possesses the preliminary detection capability for multipath signals in BeiDou navigation signals under dynamic environments.
[0036] This paper utilizes unsupervised and supervised learning to achieve early identification of multipath signals in complex urban environments. In the preliminary multipath detection of BeiDou signals in dynamic environments using synthetic machine learning, unlike other researchers who directly select a subset of satellite signal observations as the training set, this paper first removes outliers or noise values from the data samples through unsupervised clustering, effectively ensuring the accuracy of the data labels in the training set. Secondly, unlike other researchers who use unsupervised learning to obtain the signal training set, this paper combines prior knowledge of the multipath signal generation mechanism and removes outliers from the dataset through unsupervised learning, avoiding inaccuracies in data labels generated by clustering and thus increasing reliability. Through a two-stage strategy of "unsupervised cleaning followed by supervised training," first using clustering to remove outliers and then using KNN / XGBoost to learn the classification boundaries, the accuracy of the training set labels can be improved by more than 10% without adding additional sensors, thereby improving the evaluation value of the multipath signal detection model in dynamic scenarios by 8%-12%.
[0037] Specific Implementation Method 3: This implementation method further defines the method for detecting BeiDou navigation multipath signals described in Specific Implementation Method 2. In this implementation method, the two different static environments are: an open and unobstructed location on a rooftop and a location at the bottom of a tall building.
[0038] In this practical approach, BeiDou satellite signal observations were first collected in two static environments: a rooftop with unobstructed views and the base of a tall building, both largely obstructed by tall vegetation. These locations were chosen because, based on prior knowledge of multipath signals, in complex, multidimensional urban environments, multipath signal generation is typically caused by obstructions from tall vegetation, high-rise buildings, and bridge structures. Therefore, the first static scenario is rich in direct sunlight and contains relatively few multipath signals; conversely, the second static scenario is rich in multipath signals and contains relatively few direct sunlight signals. By distinguishing between these different scenarios to collect raw satellite observation datasets, a foundation is laid for subsequently constructing multipath and direct sunlight signal data separately.
[0039] Specific Implementation Method Four: This implementation method further defines the method for detecting BeiDou navigation multipath signals described in Specific Implementation Method Two. In this implementation method, the unsupervised learning method is a clustering algorithm, which is used to remove outliers from the original observation data of the direct signal and the original observation data of the multipath signal.
[0040] In this implementation, for the acquired raw signals, unsupervised learning is used to remove outliers from the original observation data, reducing the mixing of direct signals with multipath signals in the multipath signal dataset and vice versa. This makes the labels in the dataset more accurate, ensuring the subsequent learning of signal classification rules. The two signal datasets after data preprocessing are then merged. At this point, supervised learning is selected to learn the signal classification rules, that is, the merged signal dataset is shuffled and divided into a sub-training set and a sub-test set according to the appropriate proportions.
[0041] Specific Implementation Method Five: This implementation method further defines the method for detecting BeiDou navigation multipath signals described in Specific Implementation Method Two. In this implementation method, the supervised learning algorithm is the K-nearest neighbor algorithm or the extreme gradient boosting tree algorithm.
[0042] In this embodiment, the supervised learning algorithm adopts the K-Nearest Neighbors (KNN) algorithm or the Extreme Gradient Boosting (XGBoost) algorithm;
[0043] After training the signal classification and recognition model, the supervised learning algorithm with the best classification performance is selected for subsequent field testing of multipath signal detection in dynamic environments.
[0044] The KNN algorithm model is used for multipath signal detection in dynamic real-world environments. The KNN algorithm, or k-nearest neighbor algorithm, is an instance-based learning method, or a non-parametric method used for classification and regression.
[0045] The basic idea of KNN is that, given a training dataset, for a new input instance, find the k nearest neighbors in the training dataset. If the majority of these k neighbors belong to a certain class, then classify the input instance into that class. The following is the algorithm flow of KNN.
[0046] (1) Determine the K value: The K value is an important parameter of KNN and needs to be given in advance. If the K value is too small, the model will be too sensitive and easily affected by noisy data; if the K value is too large, it will include more neighbors, which may include more noise, resulting in an overly simple model.
[0047] (2) Data preparation: Prepare the training dataset. These data have labels, that is, we know the category to which each data belongs.
[0048] (3) Input the data to be predicted: Receive the data to be predicted without any category label.
[0049] (4) Calculate the distance: Use an appropriate distance metric (such as Euclidean distance) to calculate the distance between the data to be predicted and each sample in the training dataset.
[0050] (5) Select the nearest neighbor: Sort all distances and select the K training samples with the smallest distance (i.e. the K nearest neighbors of the data to be predicted).
[0051] (6) Category decision: Look at the number of times each category appears in these K samples. The category that appears most often is the category of the data to be predicted.
[0052] (7) Output results: The category determined in step 6 is used as the predicted category of the data to be predicted.
[0053] The advantages of KNN are its simplicity, ease of understanding, and lack of need for parameter estimation and training. It is also highly adaptable, suitable for both classification and regression tasks. For imbalanced datasets, the KNN algorithm can improve performance by adjusting the distance weights.
[0054] Specific Implementation Method Six: This implementation method further defines the method for detecting BeiDou navigation multipath signals described in Specific Implementation Method Two. In this implementation method, the neural network model is a feedforward neural network.
[0055] In this embodiment, the input of the feedforward neural network is the observed characteristics of the BeiDou satellite signal, and the output is the classification result of whether the signal belongs to a multipath signal. The multipath signal detection model can achieve a detection accuracy of more than 92% by using a single-layer hidden layer feedforward network. The number of parameters is only 1 / 10 of that of the existing CNN, which is suitable for implementation in the receiver DSP / FPGA with fixed-point arithmetic. The power consumption is reduced by more than 40%, which meets the low power consumption requirements of battery-powered devices such as vehicle-mounted and handheld devices.
[0056] Specific Implementation Method Seven: This implementation method further defines the method for detecting BeiDou navigation multipath signals described in Specific Implementation Method Two. In this implementation method, the observation features in the raw observation data of the direct signal and the raw observation data of the multipath signal include one or more of the following: carrier phase, pseudorange, signal-to-noise ratio, and Doppler shift.
[0057] In this embodiment, the original observations such as carrier phase, pseudorange, signal-to-noise ratio, and Doppler output by the receiver are directly reused, and multidimensional feature vectors can be constructed without additional RF channels or high sampling rate ADCs. On the same hardware platform, compared with adding a dedicated multipath suppression antenna, the BOM cost can be saved by about 15-20%.
[0058] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.
Claims
1. A method for detecting multipath signals in BeiDou navigation, characterized in that, Includes the following steps: The system receives the BeiDou navigation and positioning signal from the current target environment and uses a neural network-based multipath signal detection model to detect the received signal and identify whether the BeiDou navigation and positioning signal from the current target environment contains multipath signals.
2. The method for detecting BeiDou navigation multipath signals according to claim 1, characterized in that, The multipath signal detection model is obtained through pre-training, and the training method for the multipath signal detection model is as follows: Raw observation data of direct signals and raw observation data of multipath signals from BeiDou satellites were collected in two different static environments. Unsupervised learning methods were used to preprocess the raw observation data of the direct signal and the raw observation data of the multipath signal to remove outliers, thereby obtaining the direct signal dataset and the multipath training dataset, respectively. The direct-shoot dataset and the multipath training dataset are randomly shuffled and merged to construct the final training set; Based on the final training set, a supervised learning algorithm is used to train the neural network model to obtain a multipath signal detection model based on a neural network.
3. The method for detecting BeiDou navigation multipath signals according to claim 2, characterized in that, The two different static environments are: a rooftop location with an open and unobstructed view and a location at the bottom of a tall building.
4. The method for detecting BeiDou navigation multipath signals according to claim 2, characterized in that, The unsupervised learning method is a clustering algorithm used to remove outliers from the original observation data of the direct signal and the original observation data of the multipath signal.
5. The method for detecting BeiDou navigation multipath signals according to claim 2, characterized in that, The supervised learning algorithm is either the K-nearest neighbor algorithm or the extreme gradient boosting tree algorithm.
6. The method for detecting BeiDou navigation multipath signals according to claim 2, characterized in that, The neural network model is a feedforward neural network.
7. The method for detecting BeiDou navigation multipath signals according to claim 2, characterized in that, The observation features in the raw observation data of the direct signal and the raw observation data of the multipath signal include one or more of the following: carrier phase, pseudorange, signal-to-noise ratio, and Doppler shift.