UWB ranging error compensation method and device based on cascade residual attention network
Through the cascade residual attention network and supervised learning method, combined with the attention mechanism, the accuracy problem caused by the multipath effect in UWB ranging technology is solved, and high-precision positioning is achieved in many scenarios, suitable for low-cost wireless Positioning system.
Patent Information
- Application Number
- CN202411100713.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-12
- Publication Date
- 2024-11-08
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
UWB ranging technology is affected by multipath effect and non-sight propagation in practical applications, resulting in unstable ranging accuracy. The traditional error compensation method lacks versatility and adaptability, making it difficult to achieve high-precision positioning in various scenarios.
UWB ranging error compensation method based on cascade residual attention network is adopted, and the cascade residual network model is built through data acquisition, feature selection, and attention mechanism is added, and the distance measurement error relationship is fitted with the supervised learning method, and the Model compression, quantization and edge-end deployment to achieve fast and effective error compensation.
Fast, effective and robust UWB ranging error compensation on embedded devices with limited computing resources, improves wireless positioning accuracy and is suitable for the application needs of a variety of low-cost wireless positioning systems.
Smart Images

Figure CN118915036A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of wireless ranging, and in particular relates to a UWB ranging error compensation method and device based on a cascaded residual attention network. Background Art
[0002] Radio ranging is a distance measurement method based on electromagnetic wave application technology. Radio ranging is to measure distance by radio, which is one of the basic tasks of radio positioning and navigation. Radio ranging can be divided into three types according to its working principle: pulse ranging (also known as time ranging), phase ranging and frequency ranging. In order to meet the application requirements in fields such as high-precision position estimation and navigation, many scenarios require the ability to obtain high-precision ranging information through wireless ranging methods.
[0003] Ultra Wide Band (UWB) is an impulse radio communication technology that can achieve ranging between two different UWB devices, namely UWB tags and UWB base stations, by recording flight timestamps. Compared with traditional ranging technologies, UWB has many advantages such as strong penetration, low power consumption, good anti-multipath effect, high security, low system complexity, and centimeter-level accurate ranging capability. There are multiple channel characteristics and related parameters in the UWB chip, as well as the Channel Impulse Response (CIR) that reflects the signal propagation quality, which can be used to characterize the intrinsic relationship between the true ranging value and the ranging error value.
[0004] In theory, wireless ranging using UWB technology can obtain millimeter-level ranging information. When actually using UWB technology for wireless ranging in an ideal environment, it can usually achieve centimeter-level line-of-sight error accuracy. However, in practical applications, UWB ranging technology is often affected by multipath effects and non-line-of-sight propagation, which makes it impossible to guarantee the ranging accuracy of UWB during the measurement process. On the other hand, traditional error compensation methods usually rely on the specific characteristics of the environment and lack universality and adaptability. Summary of the invention
[0005] To solve the above problems, the present invention discloses a UWB ranging error compensation method and device based on a cascaded residual attention network, which can effectively eliminate the ranging error of UWB and is suitable for a variety of scenarios that require the application of UWB for wireless ranging, which helps to greatly improve the positioning accuracy based on UWB ranging; and can realize fast, effective and robust UWB ranging error compensation on embedded devices with limited computing resources, which is suitable for the application requirements of many low-cost wireless positioning systems.
[0006] To achieve the above object, the technical solution of the present invention is as follows: A device for UWB ranging error compensation based on a cascaded residual attention network, comprising a UWB ranging module capable of obtaining its own high-precision position coordinates and a computing module supporting edge computing; The UWB ranging module requires that the high-precision position coordinates of the UWB base station and the UWB tag are known; The UWB base station is required to be able to collect complete original CIR channel characteristics and all channel characteristics, and obtain relevant parameters of its internal registers; The computing module supporting edge computing meets the minimum configuration requirements for deploying the UWB ranging error compensation method based on the cascaded residual attention network.
[0007] The data set required by the ranging error compensation method can be completely collected by the UWB tag and the UWB base station.
[0008] A UWB ranging error compensation method based on a cascaded residual attention network comprises the following steps: Step 1: data collection and feature selection; Step 2: Build a cascaded residual network model, add an attention mechanism and implement UWB ranging error compensation; Step 3: compress, quantize and deploy the model on the edge, and implement the model deployment on a computing module that supports edge computing.
[0009] Furthermore, the data collection described in step 1 specifically includes UWB ranging values, UWB channel characteristics, complete original CIR signals, and the actual distance between the UWB base station and the tag.
[0010] The feature selection is used to screen out key CIR signal features and key channel features from the original CIR signal and all channel features collected by the UWB ranging module to form an input data set; The feature selection uses a feature extraction module to perform feature extraction and reduce the input data dimension of the residual attention module.
[0011] Using feature selection to screen out key CIR data and key channel features is highly relevant to the UWB ranging error compensation task.
[0012] Furthermore, the cascade residual network model described in step 2 is constructed, and the complex implicit relationship between the input data set and the UWB ranging error is fitted using a supervised learning method, and an input module of the input data set obtained by feature selection is constructed through a linear input transformation; The input of the cascade residual network model is an input data set obtained by a feature selection method, and the output is a UWB ranging error compensation result; the model includes an input layer, a feature extraction module, a cascade structure, a residual network layer and an output layer.
[0013] The cascade structure can connect multiple residual blocks, each residual block contains a standard residual connection. The cascade structure can enable the network to comprehensively consider the mutual influence between features at a deep level and prevent information loss during training. The attention mechanism is used to re-evaluate and effectively learn the ranging error fitting of the effective features extracted by the cascade residual network model from the input data set obtained by feature selection and the representation after dimensionality reduction, highlight the features important to the target output, weaken the useless features or the features with low correlation, and improve the performance of the neural network model; enable the network to adaptively focus on processing the features that have the greatest impact on the final ranging result.
[0014] The method of adding an attention mechanism and realizing UWB ranging error compensation firstly adds an attention mechanism to the cascaded residual network, wherein each residual block contains a standard residual connection and at least one attention unit, and these attention units are responsible for enhancing the model's recognition ability for key features; secondly, the model weights are trained by a neural network supervised training method; wherein the training label is the error value between the real distance in the input data set and the UWB measured distance.
[0015] Furthermore, the model compression, quantization and edge deployment described in step 3 are implemented, and the model deployment is implemented on a computing module that supports edge computing, for volume compression and weight quantization of the model with UWB ranging error compensation, and deployment of the model on a computing module that supports edge computing; Among them, model compression uses volume compression to compress the original neural network model into a TFlite file to reduce the volume of the original model, making it more conducive to deployment on a computing module that supports edge computing; Quantization is weight quantization, and the TFlite file of the compressed model is fully integer quantized, making it more conducive to online reasoning on the computing module that supports edge computing and further reducing the model size; Use edge deployment to convert the compressed and weight-quantized TFlite file into C / C++ source code and burn it to the computing module that supports edge computing to complete the deployment.
[0016] The working principle of the present invention is: After using the UWB ranging module to collect complete CIR data and all channel features, the key CIR data and channel features are selected through feature selection to form an input data set. Then, a cascaded residual network model is built and the attention mechanism is added for model training. The effective channel features obtained by the UWB ranging module and the signal features of the CIR after dimensionality reduction in the input data set obtained by feature selection are re-evaluated and effectively fitted to learn the ranging error, so that it has the ability to effectively predict the UWB ranging error. Finally, through compression, quantization and edge deployment, the trained model is used on the computing module that supports edge computing to realize the UWB ranging error prediction, and the distance error between the UWB tag and the UWB base station is compensated and eliminated.
[0017] The beneficial effects of the present invention are: The present invention proposes a UWB ranging error compensation method based on a cascaded residual attention network, which realizes UWB ranging error compensation through a cascaded residual network and an attention mechanism. The wireless ranging technology can be applied to the UWB ranging module to collect complete original CIR channel features and all channel features, and obtain relevant parameters of its internal registers. The key CIR signal features and key channel features are screened out from the original CIR signal and all channel features collected by the UWB ranging module by feature selection to form an input data set. Compared with the use of complete original CIR information, the feature selection method greatly reduces the dimension of the input data set, and does not lose the signal features closely related to the UWB ranging error value. The supervised learning method is used to fit the complex implicit relationship between the input data set and the UWB ranging error, and the input module of the input data set obtained by feature selection is constructed by linear input transformation. The residual connection can avoid the gradient vanishing problem in the deep network training process. By cascading multiple residual blocks, the network comprehensively considers the mutual influence between features at a deep level, prevents information loss during the training process, and improves the effectiveness of feature extraction. By adding the attention mechanism, the effective features extracted from the input data set obtained by feature selection by the cascade residual network model and the representation after dimensionality reduction are re-evaluated and effectively fitted with the ranging error, the features important to the target output are highlighted, and the useless features or features with low correlation are weakened, so as to improve the performance of the neural network model, so that the network can adaptively focus on the features that have the greatest impact on the final ranging result. The cascade residual attention network model is trained by the supervised training method of the neural network. The input of the model is the input data set obtained by the feature selection method, and the output is the UWB ranging error compensation result. Finally, through compression, quantization and edge deployment, the trained model is used to realize UWB ranging error prediction on the computing module supporting edge computing. The present invention can realize fast, effective and robust UWB ranging error compensation on embedded devices with limited computing resources, which is suitable for the application requirements of many low-cost wireless positioning systems and helps to further improve the positioning accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] Figure 1 This is an overall schematic diagram of the UWB ranging error compensation method proposed in the present invention.
[0019] Figure 2 The figure is a schematic diagram of the overall flow of the UWB ranging error compensation method proposed in the present invention.
[0020] Figure 3 This is a schematic diagram of obtaining a 131-dimensional input data set by using feature selection in step 1 of the present invention.
[0021] Figure 4 Schematic diagram of the neural network model structure in step 2 of the present invention.
[0022] Figure 5 It is an overall schematic diagram of step 2 of the present invention.
[0023] Figure 6 It is an overall schematic diagram of step 3 of the present invention. DETAILED DESCRIPTION
[0024] The present invention will be further explained below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention.
[0025] As shown in the figure, the UWB ranging error compensation method based on the cascaded residual attention network described in the present invention includes the following steps: 1) data collection and feature selection; 2) building a cascaded residual network model, adding an attention mechanism and realizing UWB ranging error compensation; 3) model compression, quantization and edge deployment, and realizing the deployment of the model on a computing module that supports edge computing.
[0026] The present invention also introduces a related device, including a UWB ranging module capable of obtaining its own high-precision position coordinates and a computing module supporting edge computing; The UWB ranging module requires that the high-precision position coordinates of the UWB base station and the UWB tag are known; The UWB base station is required to be able to collect complete original CIR channel characteristics and all channel characteristics, and obtain relevant parameters of its internal registers; The computing module supporting edge computing meets the minimum configuration requirements for the deployment of the UWB ranging error compensation method based on the cascaded residual attention network; The data set required by the distance measurement error compensation method can be completely collected by the UWB tag and the UWB base station; The data set includes, UWB ranging value, UWB channel characteristics, complete original CIR signal and the real distance between the UWB base station and the tag; The feature selection described in step 1) is used to select key CIR signal features and key channel features from the original CIR signal and all channel features collected by the UWB ranging module to form an input data set; The feature selection uses a feature extraction module to perform feature extraction and reduce the input data dimension of the residual attention module.
[0027] Step 2) includes 5 steps: a: Build input module; b: Feature extraction module reduces the input dimension; c: Construct cascade residual connections; d: Add attention mechanism; e: Neural network supervised training.
[0028] The input of the cascade residual network model is an input data set obtained by a feature selection method, and the output is a UWB ranging error compensation result; the model includes an input layer, a feature extraction module, a cascade structure, a residual network layer and an output layer.
[0029] The cascade structure can connect multiple residual blocks, each residual block contains a standard residual connection. The cascade structure can enable the network to comprehensively consider the mutual influence between features at a deep level and prevent information loss during training. The attention mechanism is used to re-evaluate and effectively learn the ranging error fitting of the effective features extracted by the cascade residual network model from the input data set obtained by feature selection and the representation after dimensionality reduction, highlight the features important to the target output, weaken the useless features or the features with low relevance, and improve the performance of the neural network model; The method of adding an attention mechanism and realizing UWB ranging error compensation firstly adds an attention mechanism to the cascaded residual network, wherein each residual block contains a standard residual connection and at least one attention unit, and these attention units are responsible for enhancing the model's recognition ability for key features; secondly, the model weights are trained by a neural network supervised training method; wherein the training label is the error value between the real distance in the input data set and the UWB measured distance.
[0030] The model compression, quantization and edge deployment described in step 3) are implemented, and the model deployment is implemented on a computing module that supports edge computing, which is used to perform volume compression and weight quantization on the model with UWB ranging error compensation, and deploy it on a computing module that supports edge computing.
[0031] The present invention proposes a UWB ranging error compensation method based on a cascaded residual attention network, which realizes UWB ranging error compensation through a cascaded residual network and an attention mechanism. The wireless ranging technology can be applied to the UWB ranging module to collect complete original CIR channel features and all channel features, and obtain relevant parameters of its internal registers. The key CIR signal features and key channel features are screened out from the original CIR signal and all channel features collected by the UWB ranging module by feature selection to form an input data set. Compared with the use of complete original CIR information, the feature selection method greatly reduces the dimension of the input data set, and does not lose the signal features closely related to the UWB ranging error value. The supervised learning method is used to fit the complex implicit relationship between the input data set and the UWB ranging error, and the input module of the input data set obtained by feature selection is constructed by linear input transformation. The residual connection can avoid the gradient vanishing problem in the deep network training process. By cascading multiple residual blocks, the network comprehensively considers the mutual influence between features at a deep level, prevents information loss during the training process, and improves the effectiveness of feature extraction. By adding the attention mechanism, the effective features extracted from the input data set obtained by feature selection by the cascade residual network model and the representation after dimensionality reduction are re-evaluated and effectively learned by fitting the ranging error, the features important to the target output are highlighted, the useless features or the features with low correlation are weakened, and the performance of the neural network model is improved, so that the network can adaptively focus on the features that have the greatest impact on the final ranging result. The cascade residual attention network model is trained by the supervised training method of the neural network. The input of the model is the input data set obtained by the feature selection method, and the output is the UWB ranging error compensation result. Finally, through compression, quantization and edge deployment, the trained model is used on the computing module supporting edge computing to realize UWB ranging error prediction, and the distance error between the UWB tag and the UWB base station is compensated and eliminated. The present invention can realize fast, effective and robust UWB ranging error compensation on embedded devices with limited computing resources, which is suitable for the application requirements of many low-cost wireless positioning systems and helps to further improve the positioning accuracy.
[0032] In summary, the UWB ranging error compensation method and device based on the cascaded residual attention network proposed in the present invention can effectively eliminate the UWB ranging error, and can be deployed on embedded devices with limited computing resources to achieve fast, effective and robust UWB ranging error compensation. It is suitable for a variety of scenarios that require the use of UWB for wireless ranging, and helps to greatly improve the ranging and positioning accuracy based on UWB.
[0033] It should be noted that the above content only illustrates the technical idea of the present invention and cannot be used to limit the protection scope of the present invention. For ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications all fall within the protection scope of the claims of the present invention.
Claims
1. A device for UWB ranging error compensation based on a cascaded residual attention network, characterized in that: It includes a UWB ranging module that can obtain its own high-precision position coordinates and a computing module that supports edge computing; The UWB ranging module requires that the high-precision position coordinates of the UWB base station and the UWB tag are known; The UWB base station is required to be able to collect complete original CIR channel characteristics and all channel characteristics, and obtain relevant parameters of its internal registers; The computing module supporting edge computing meets the minimum configuration requirements for the deployment of the UWB ranging error compensation method based on the cascaded residual attention network; The data set required by the ranging error compensation method is completely collected by the UWB tag and the UWB base station.
2. A UWB ranging error compensation method based on a cascaded residual attention network using the device of claim 1, characterized in that: The following steps are involved: Step 1, data collection and feature selection; Step 2: Build a cascaded residual network model, add an attention mechanism and implement UWB ranging error compensation; Step 3: compress, quantize and deploy the model on the edge, and implement the model deployment on a computing module that supports edge computing.
3. The UWB ranging error compensation method based on cascaded residual attention network according to claim 2 is characterized in that: The data collection described in step 1 specifically includes UWB ranging value, UWB channel characteristics, complete original CIR signal and the real distance between the UWB base station and the tag; The feature selection is used to screen out key CIR signal features and key channel features from the original CIR signal and all channel features collected by the UWB ranging module to form an input data set; The feature selection uses a feature extraction module to perform feature extraction and reduce the input data dimension of the residual attention module.
4. The UWB ranging error compensation method based on cascaded residual attention network according to claim 2, characterized in that: The construction of the cascaded residual network model described in step 2 is to fit the complex implicit relationship between the input data set and the UWB ranging error using a supervised learning method, and to construct an input module of the input data set obtained by feature selection through a linear input transformation; The input of the cascade residual network model is an input data set obtained by a feature selection method, and the output is a UWB ranging error compensation result; the model includes an input layer, a feature extraction module, a cascade structure, a residual network layer and an output layer, and the training label is the error value between the real distance and the UWB distance measurement; wherein the real distance is calculated by the high-precision position coordinates of both the UWB tag and the UWB base station; The cascade structure can connect multiple residual blocks, each residual block contains a standard residual connection. The cascade structure can enable the network to comprehensively consider the mutual influence between features at a deep level and prevent information loss during training. The attention mechanism is used to extract effective features and representations after dimensionality reduction from the input data set obtained by feature selection for the cascade residual network model, re-evaluate and effectively learn the ranging error fitting, highlight the features important to the target output, weaken the useless features or the features with low correlation, and improve the performance of the neural network model; enable the network to adaptively focus on processing the features that have the greatest impact on the final ranging result; The method of adding an attention mechanism and realizing UWB ranging error compensation firstly adds an attention mechanism to the cascaded residual network, wherein each residual block contains a standard residual connection and at least one attention unit, and these attention units are responsible for enhancing the model's recognition ability for key features; secondly, the model weights are trained by a neural network supervised training method; wherein the training label is the error value between the real distance in the input data set and the UWB measured distance.
5. The UWB ranging error compensation method based on cascaded residual attention network according to claim 2, characterized in that: The model compression, quantization and edge deployment described in step 3, and the deployment of the model on the computing module supporting edge computing, are used to compress the volume and quantize the weight of the model with UWB ranging error compensation, and deploy it on the computing module supporting edge computing; Among them, model compression uses volume compression to compress the original neural network model into a TFlite file to reduce the volume of the original model, making it more conducive to deployment on a computing module that supports edge computing; Quantization is weight quantization, and the TFlite file of the compressed model is fully integer quantized, making it more conducive to online reasoning on the computing module that supports edge computing and further reducing the model size; Use edge deployment to convert the compressed and weight-quantized TFlite file into C / C++ source code and burn it to the computing module that supports edge computing to complete the deployment.
6. The UWB ranging error compensation method based on cascaded residual attention network according to claim 2, characterized in that: Its working principle is: After using the UWB ranging module to collect complete CIR data and all channel characteristics, the key CIR data and channel characteristics are selected through feature selection to form an input data set; then a cascaded residual network model is built and the attention mechanism is added for model training. The effective channel characteristics obtained by the UWB ranging module and the signal characteristics of the CIR after dimensionality reduction in the input data set obtained by feature selection are re-evaluated and effectively fitted to learn the ranging error, so that it has the effective prediction ability of UWB ranging error; finally, through compression, quantization and edge deployment, the trained model is used on the computing module that supports edge computing to realize UWB ranging error prediction, and the distance error between the UWB tag and the UWB base station is compensated and eliminated.
Citation Information
Patent Citations
UWB ranging error compensation algorithm and device based on attention mechanism and transfer learning
CN115329675A
Distance measurement error compensation method and system for ship construction workshop personnel positioning
CN115802485A
UWB NLOS propagation error suppression method based on deep learning
CN116321423A
UWB line-of-sight / non-line-of-sight identification equipment and algorithm
CN117851881A
LoRa ranging error suppression positioning method based on deep learning
CN118283788A
Cited By
UWB / INS adaptive positioning method and system based on deep learning
CN120368979A
Self-supervised ultra-wideband positioning method based on improved double-delay depth deterministic strategy
CN121099418A
Self-supervised ultra-wideband positioning method based on improved double-delay deep deterministic policy
CN121099418B
UWB ranging error compensation algorithm and device based on channel state dynamic routing
CN121218096A
Indoor UWB (Ultra Wideband) multi-mode ad hoc network credible positioning system for ship
CN122430788A
