CN-121763206-B - Multi-dimensional electromagnetic field diagram-based wireless array single-base station positioning method and system

CN121763206BCN 121763206 BCN121763206 BCN 121763206BCN-121763206-B

Abstract

The invention discloses a wireless array single-base station positioning method and system based on a multidimensional electromagnetic field diagram, and belongs to the technical field of indoor/underground space positioning. The method comprises the steps of constructing an electromagnetic field diagram containing time domain, frequency domain and space domain multidimensional features by utilizing a high-precision track acquisition means to cooperate with an array single base station, constructing a local relative coordinate system of the base station by utilizing a vector orientation method, converting absolute coordinates into relative coordinates to realize decoupling of a model and a geographic environment, performing geometric projection pretreatment based on a current scene height difference, eliminating installation height influence, introducing domain marks into input features to adapt to different moving targets, constructing a TCN-transducer mixed model fusing a channel attention mechanism for training, and finally restoring the coordinates through a reverse transformation matrix in a new scene and supporting rapid fine adjustment of a small amount of data. The invention realizes the positioning with high precision, high efficiency and strong generalization capability under a single base station.

Inventors

DU SHITONG
WEI BAOGUO
LI YIFAN
YANG ZIHAN
LI SHUANG
HUANG LU
CHENG JIANQIANG
LIANG XIAOHU
CHEN CHONG

Assignees

中国电子科技集团公司第五十四研究所

Dates

Publication Date: 20260505
Application Date: 20260302

Claims (7)

1. A wireless array single base station generalization positioning method based on a multidimensional electromagnetic field diagram is characterized by comprising the following steps: Step 1, binding a positioning terminal with track acquisition equipment, continuously moving in a region to be detected to acquire a space track map of the positioning terminal, acquiring multi-dimensional radio frequency signal characteristics covering a time domain, a frequency domain and a space domain by utilizing an array single base station, and constructing an electromagnetic field map containing dynamic characteristics by the multi-dimensional radio frequency signal characteristics and the space track map through time alignment; Step 2, constructing a local coordinate system with an array single base station as an origin by using a vector orientation method based on the acquired electromagnetic field diagram, calculating conversion parameters from an absolute coordinate system to the local coordinate system, and converting the absolute space position in the electromagnetic field diagram constructed in the step 1 into a relative coordinate position, wherein the absolute coordinate system is a track acquisition equipment coordinate system; Step 3, acquiring the height difference between the base station and the positioning terminal of the current scene, and performing geometric dimension reduction projection on the ranging value; step 4, constructing a hybrid neural network model comprising a time convolution network of a channel attention mechanism and a transducer encoder, training the model by utilizing the electromagnetic field map data processed in the step 2 and the explicit domain mark in the step 3, and learning the mapping relation between multidimensional signal characteristics and local relative coordinates; and 5, measuring absolute coordinates and orientation reference points of a new base station in the positioning process of the new environment, calculating an inverse transformation matrix from the absolute coordinate system to a local coordinate system in the new environment, acquiring electromagnetic field map data of the new environment if the accuracy of the pre-training model is reduced due to environmental differences, freezing TCN layer parameters of physical characteristics at the front end of the model, thawing a transducer and an output layer at the rear end, performing quick fine tuning training on the pre-processing model by using the new electromagnetic field map data, outputting relative coordinates of the model after fine tuning training during real-time positioning, and calculating the absolute position through inverse transformation.
2. The method for generalizing and positioning a wireless array single base station based on a multi-dimensional electromagnetic field map according to claim 1, wherein the specific configuration of the multi-dimensional radio frequency signal characteristics in step 1 comprises time domain characteristics including ranging observations calculated based on time of flight and channel impulse responses characterizing signal energy distribution with time delay; the frequency domain features comprise carrier phase observation values and channel frequency response which characterizes fading characteristics of signals on different frequency components; The space domain features comprise a phase difference matrix among antenna units in the array receiving device, wherein the phase difference matrix represents the arrival angle space distribution information of signals.
3. The method for generalizing and positioning the wireless array single base station based on the multidimensional electromagnetic field map according to claim 2, wherein the specific mode of the step 2 is as follows: Acquiring absolute coordinates of a base station center Selecting any point in the X-axis positive direction under the base station coordinate system as a directional reference point to obtain the absolute coordinate of the point Calculating a rotation angle: ; rotation matrix of absolute coordinate system to local coordinate system Translation vector Expressed as: , , For each piece of training data in the electromagnetic field diagram, the space position contained in the training data is calculated from absolute coordinates Conversion into base station local relative coordinates Decoupling of model training and a coordinate system is realized; Wherein, the 。
4. The method for generalizing and positioning a single base station of a wireless array based on a multi-dimensional electromagnetic field map according to claim 3, wherein the specific mode of step 3 is that physical prior is introduced in a feature extraction stage aiming at the problem of uncertainty of base station installation height and carrier height, wherein the base station height is The height of the positioning terminal is The height difference is expressed as: , For the original ranging value Performing geometric dimension reduction projection to obtain a projected plane distance value The method comprises the following steps: , aiming at a mixed positioning scene of a person and a vehicle, a one-dimensional explicit domain marking scalar is added in an input feature vector Obtaining the feature vector of the next deep learning input : , Wherein, the Respectively represent any time steps The time array base station is related to the reduced-dimension distance characteristic, the array phase, the channel impulse response and the domain self-adaptive zone bit.
5. The wireless array single base station generalization positioning method based on the multi-dimensional electromagnetic field diagram, which is characterized by comprising the following specific modes of step 4, constructing a deep neural network, wherein a feature extraction layer adopts a stacked residual time convolution network, embedding SE-Block in each residual Block, automatically inhibiting antenna signals seriously interfered by multipath by learning importance weights of antenna channels, a time sequence modeling layer adopts a transducer encoder, and captures long-distance time sequence dependence by utilizing a self-attention mechanism, outputs layer regression prediction to two-dimensional local coordinates of a base station, iteratively optimizes an objective function based on a AdamW optimizer and SmoothL loss function, and introduces a random mask strategy in training to simulate signal packet loss conditions so as to improve robustness; the feature extraction layer is used for extracting local features by using the expansion causal convolution and is used for inputting sequences Convolution kernel And an expansion ratio of Is defined as: ; To suppress multipath interference, SE-Block is introduced to calculate channel weight vector And (3) recalibrating the feature map: ; Wherein, the Representing an intermediate feature tensor of the expanded convolved output, which contains the unweighted multi-channel timing features, Representing the attention weighted enhancement feature tensor; The time sequence modeling layer introduces position coding PE to reserve sequence order information, calculates global time sequence association by utilizing a multi-head self-attention mechanism and outputs context vectors; objective function, constructing an optimization objective by adopting a smooth L1 loss function: , Wherein the method comprises the steps of The method is characterized in that the method is represented as square loss when the error is small, and is represented as absolute value loss when the error is large, so that the robustness of the system is enhanced; representing the total number of samples for the current training batch, Representing the base station local relative coordinate value of the model for the i sample prediction; And the local relative coordinate value of the real base station corresponding to the ith sample is represented.
6. The wireless array single base station generalization positioning method based on the multi-dimensional electromagnetic field map is characterized in that the specific mode of performing fine tuning training on a model part network layer in the step 5 is that electromagnetic field map data under a new scene is collected, the same preprocessing is realized by repeating the step 2 and the step3, a time convolution network module for extracting the characteristics of a bottom layer physical signal and weight parameters of SE-Block in a mixed neural network model are frozen, the weight parameters of a thawing transducer encoder module and an output layer are thawed, and the parameters of the thawing layer are iteratively updated at a learning rate lower than that of initial training by utilizing the new scene electromagnetic field map data, so that the model is adapted to the multipath distribution characteristics of a new environment; The freezing stage, locking the front-end parameters so that their gradients do not participate in counter-propagation, i.e Reserving general physical layer signal extraction capability; the updating stage is to apply gradient descent updating to the back-end parameters: ; Wherein, the Representing base parameters/front-end parameters, the part of the network being responsible for extracting the underlying physical characteristics of the signal; As the back-end parameters, the network is responsible for mapping abstract signal features to specific geometric space coordinates, and retraining is needed due to the change of the multipath reflection structure of the new environment; For the purpose of the gradient operator, To fine tune the learning rate, 1/10 of the initial training learning rate is set; finally, using the formula And restoring the relative coordinates output by the model into absolute coordinates in the new scene.
7. A wireless array single base station generalization positioning system for realizing the wireless array single base station generalization positioning method based on the multi-dimensional electromagnetic field map according to any one of claims 1 to 6, which is characterized by comprising a signal acquisition module, a track acquisition module, a data preprocessing module, a model training and fine tuning module and a coordinate post-processing module; The system comprises a signal acquisition module, a track acquisition module, a data preprocessing module, a model training module, a fine tuning module and a coordinate post-processing module, wherein the signal acquisition module is used for completing the acquisition task of multidimensional radio frequency signal characteristics covering time domain, frequency domain and space domain in the step 1, the track acquisition module is used for completing the track acquisition in the step 1, the data preprocessing module is used for decoupling and standardization of a coordinate system in the step 2 and the step 3, the model training module is used for completing the related task in the step 4, the fine tuning module is used for carrying out fine tuning training on a pre-training model in the step 5, and the coordinate post-processing module is used for converting a relative coordinate into an absolute position through inverse transformation solution.

Description

Multi-dimensional electromagnetic field diagram-based wireless array single-base station positioning method and system Technical Field The invention belongs to the technical field of indoor/underground space positioning, and particularly relates to a wireless array single-base station positioning method and system based on a multidimensional electromagnetic field diagram, which are suitable for wireless positioning systems with array receiving capability, such as Ultra Wideband (UWB), pseudolites, 5G and the like. Background Currently, with the rapid development of the internet of things, intelligent manufacturing and automatic driving technologies, high-precision positioning of indoor and underground spaces plays an increasingly important role in the fields of warehouse logistics, personnel management, robot navigation and the like. Because satellite signals cannot be covered indoors, a positioning technology based on radio signals (such as ultra wideband UWB, pseudolites, 5G and the like) becomes a core support for solving the positioning problem of the last kilometer, and has important social value and economic significance. Conventional wireless positioning systems rely mainly on geometric measurement principles, but in practical applications, challenges are faced in that multiple base station joint solutions (such as TDOA) require complex clock synchronization and high hardware deployment costs. Conventional single-base station geometry solution (aoa+ranging) suffers from serious distortion of signal phase and time-of-flight in non-line-of-sight (NLOS) and rich multipath environments, resulting in dramatic degradation of positioning performance. To overcome the limitations of geometric methods, fingerprint localization techniques are widely studied. However, the existing fingerprint positioning method has the following bottlenecks that firstly, the construction efficiency is low, the traditional static grid acquisition method is time-consuming and labor-consuming, the traditional static grid acquisition method is difficult to land on a large scale scene, secondly, the dynamic adaptability is poor, the static fingerprints cannot reflect Doppler effect and dynamic shielding characteristics when targets move, thirdly, the generalization capability is weak, the existing model is usually strongly coupled with absolute coordinates of specific scenes and the base station installation height, once the environment changes or the base station position changes, the model is invalid, fourthly, the multi-target adaptability is poor, the motion characteristics of pedestrians and vehicles and the shielding modes of signals are quite different, and the traditional single model is difficult to simultaneously consider. Disclosure of Invention In order to solve the technical problems mentioned in the background art, the invention provides a wireless array single-base station positioning method and system based on a multidimensional electromagnetic field diagram. The dynamic automatic construction of the electromagnetic field diagram is realized through a laser SLAM technology, decoupling of a coordinate system and a physical environment is realized through a vector orientation method and a geometric height projection technology, a domain self-adaptive mark is introduced into a TCN-transducer network of a fusion channel attention mechanism, and finally, the positioning of high precision, high efficiency and strong generalization capability under a single base station is realized through a self-adaptive fine tuning strategy under a new environment. In order to achieve the above purpose, the technical scheme adopted by the invention is as follows: a wireless array single base station generalization positioning method based on a multidimensional electromagnetic field diagram comprises the following steps: Step 1, binding a positioning terminal with track acquisition equipment, continuously moving in a region to be detected to acquire a space track map of the positioning terminal, acquiring multi-dimensional radio frequency signal characteristics covering a time domain, a frequency domain and a space domain by utilizing an array single base station, and constructing an electromagnetic field map containing dynamic characteristics by the multi-dimensional radio frequency signal characteristics and the space track map through time alignment; Step 2, constructing a local coordinate system with an array single base station as an origin by using a vector orientation method based on the acquired electromagnetic field diagram, calculating conversion parameters from an absolute coordinate system to the local coordinate system, and converting the absolute space position in the electromagnetic field diagram constructed in the step 1 into a relative coordinate position, wherein the absolute coordinate system is a track acquisition equipment coordinate system; Step 3, acquiring the height difference between the base station and the positi