Multi-dimensional joint closed-loop optimization positioning method based on channel multi-domain characteristic digital twinning

Abstract

The invention discloses a multi-dimensional joint closed-loop optimization positioning method based on channel multi-domain characteristic digital twinning, and aims to improve the positioning precision in multi-path propagation and non-line-of-sight propagation environments. A multi-path received signal calculation method is given by analyzing the components of a multi-path signal and considering the time delay, attenuation and phase change of each path. Multi-domain decomposition is carried out on a received signal by adopting wavelet transform, features are extracted from a time domain, a frequency domain and a Doppler domain, and a waveform difference degree calculation method based on multi-domain features is defined. A multi-base-station signal fusion positioning mathematical model based on channel multi-domain characteristic digital twinning is constructed, and compared with an existing researched single-base-station closed-loop positioning method, the multi-dimensional joint closed-loop positioning is realized by arranging a plurality of base stations in the physical world. A posterior probability is deduced by combining the comprehensive difference degree of a plurality of base stations with Bayesian inference, in posterior probability optimization, a dynamic optimization mechanism is designed, Newton method rapid iterative optimization is proposed according to a hardware resource selection optimization algorithm and a low-computing-performance environment, and optimization is realized by adopting a grid search method in a high-computing-performance environment. And accurate estimation of the position of the to-be-positioned node is realized.

Description

Technical Field

[0001] The present invention relates to a multi-dimensional joint closed-loop optimization positioning method based on channel multi-domain feature digital twins, which is suitable for positioning needs in multi-path propagation, signal attenuation and high noise environments, and is used to improve the positioning accuracy of traditional positioning methods in complex environments. Background Art

[0002] In industrial environments, the widespread application of the Internet of Things is leading the transformation of manufacturing and positioning technology. Whether it is production automation, warehouse logistics, or equipment monitoring, accurate positioning technology plays a key role in its intelligent operation. However, in wireless sensor networks and complex industrial environments, positioning technology faces problems such as multipath propagation (MPCs) and non-line-of-sight (NLOS), which not only limit the accuracy of traditional positioning systems, but also limit the real-time and stability of positioning. Multipath propagation is a basic phenomenon in wireless signal propagation, which is caused by the reflection, refraction and scattering of signals on walls, metal equipment and other obstacles. This propagation phenomenon causes the signal to pass through multiple paths when it reaches the receiving node, resulting in constructive or destructive interference, changing the amplitude and phase of the signal, thereby affecting the positioning accuracy. In addition, non-line-of-sight propagation further exacerbates the complexity of positioning. When the direct line-of-sight path between the node to be located and the base station is blocked, the signal propagation path becomes longer and the signal quality decreases, especially when the number of reference nodes is limited, which may lead to connection interruption or positioning failure.

[0003] In response to the multipath propagation problem, existing research has mainly focused on suppressing multipath interference and effectively utilizing multipath information. In terms of interference suppression, methods include using dual-polarized antennas to distinguish polarization patterns to reduce the impact of indirect path components (IPCs), using particle filters to dynamically separate direct path components (DPCs) from IPCs, and reducing signal deviations through frequency hopping and multi-carrier technology. These technologies effectively improve the interference of multipath on positioning accuracy, but usually rely on complex hardware equipment or high computing costs. At the same time, research on using multipath information to improve positioning accuracy has also made progress. Wideband and ultra-wideband technologies can resolve multipath signals in the time domain and regard the reflection path as a virtual anchor point. Millimeter wave technology combines its high resolution and multi-antenna array advantages to resolve multipath characteristics in the time and angle domains. These methods significantly improve the robustness of positioning systems by converting multipath into usable information. However, in scenarios with severe signal overlap or complex environments, these technologies still face challenges in terms of multipath resolution efficiency and stability.

[0004] Non-line-of-sight propagation causes signal distortion, which limits the performance of the positioning system. Related research mainly focuses on NLOS identification, error compensation and multi-sensor fusion. In terms of identification, machine learning methods (such as convolutional neural networks and adaptive boosting algorithms) can effectively classify LOS and NLOS conditions by analyzing channel impulse responses or ranging characteristics. However, accuracy in dynamic environments and the generalization ability of models are still limiting factors. In terms of error compensation, convex optimization and sparse estimation methods are used to model and correct NLOS errors. Although these methods have significantly improved accuracy, they have high computational complexity and are difficult to meet real-time requirements. In addition, multi-sensor fusion and the use of environmental features provide a more robust solution to the NLOS problem. For example, the combination of inertial navigation and ultra-wideband signals significantly improves positioning accuracy, and the fusion of visual sensor-assisted channel information also shows superior performance in complex scenarios. However, these methods often have high hardware requirements and require a lot of computing resources during the fusion process.

[0005] In summary, existing methods still have problems such as insufficient multipath resolution efficiency, poor generalization ability and high computing resource requirements when dealing with multipath and non-line-of-sight propagation. These problems limit the application of positioning systems in complex industrial environments.

[0006] In order to solve the challenges brought by the above-mentioned multipath interference and non-line-of-sight propagation, the present invention proposes a multi-dimensional joint closed-loop optimization positioning method based on channel multi-domain feature digital twins. By analyzing the components of the multipath signal, considering the delay, attenuation and phase change of each path, the formula for the multipath received signal is given. By constructing a virtual digital twin model of the multi-base station positioning scene in the physical world, the wavelet transform is used to decompose the signal in multiple domains, extract the characteristics of the time domain, frequency domain and Doppler domain, and calculate the multi-domain comprehensive difference of each base station signal. The signal difference of multiple base stations is used as the core indicator to measure the degree of matching between virtual and actual signals. Combined with the Bayesian inference method, the posterior estimate of the position of the node to be located is obtained. In order to further optimize the position estimation, a dynamic optimization posterior probability position estimation method is designed, which can flexibly select the optimal optimization strategy according to the hardware resource conditions. In a low-performance environment, the Newton method is used for rapid iterative optimization; in a high-performance environment, the grid search method is used for precise optimization, thereby achieving accurate estimation of the position of the virtual node to be located. Summary of the invention

[0007] The present invention provides a multi-dimensional joint closed-loop optimization positioning method based on channel multi-domain feature digital twins, comprising the following steps:

[0008] Step 1: Calculation of wavelet transform difference based on multi-domain feature decomposition

[0009] The present invention analyzes the components of multipath signals, considers the delay, attenuation and phase change of each path, and gives a formula for multipath received signals. By constructing a channel digital twin system, the signal characteristics of different paths are extracted to generate simulation waveforms. The received physical signal and virtual signal are decomposed into multiple domains using wavelet transform, and the delay characteristics of the time domain, the attenuation characteristics of the frequency domain and the phase change characteristics of the Doppler domain are extracted to obtain the wavelet coefficients of the multiple domains. The difference of the wavelet coefficients is used as the multi-domain difference of the signal to construct a likelihood function based on the multi-domain difference.

[0010] Step 2: Construction of a multi-dimensional joint positioning mathematical model based on channel multi-domain feature digital twins

[0011] The present invention constructs a mathematical model for the multi-dimensional joint closed-loop optimization positioning method of the digital twin of the multi-domain characteristics of the channel. Compared with the single base station closed-loop positioning method that has been studied, the present invention constructs a multi-dimensional joint closed-loop optimization positioning model by deploying multiple base stations in the physical world and mapping them to the virtual world. The Gaussian distribution is used to represent the likelihood function of the multi-domain difference, and the joint effect of multiple base stations is taken into account to form a multi-dimensional joint likelihood function based on the multi-domain difference. The prior estimate of the node to be located is set, and the Bayesian inference method is used to derive the posterior probability based on the channel digital twin, and a mathematical model based on multi-base station signal fusion positioning is constructed.

[0012] Step 3: Maximum a posteriori probability position estimation method based on dynamic optimization

[0013] In order to meet the hardware environments with different computing capabilities, the present invention designs a dynamic optimization method for maximum a posteriori probability position estimation. When the positioning system runs in a low-performance computing environment, the present invention uses the negative logarithm of the posterior estimate as the loss function, derives the gradient and Hessian matrix of the loss function in the iterative process, and realizes the rapid iteration of position estimation through Newton's method. When the system runs in a high-performance computing environment, the search range and step size are set, and the maximum a posteriori probability position estimation is realized by dividing the grid and traversing each grid point to calculate the posterior probability. By dynamically selecting the optimization algorithm, the positioning system can automatically adjust the algorithm according to the hardware resources, taking into account both computing efficiency and positioning accuracy.

[0014] Compared with the prior art, the advantages of the present invention are:

[0015] 1. The present invention performs multi-domain decomposition of the signal received by the base station in the positioning model through wavelet transform, considers the key features of the time domain, frequency domain and Doppler domain, and defines a waveform difference calculation method based on multi-domain features based on the indicators of delay, attenuation and phase change.

[0016] 2. Compared with the closed-loop optimization positioning method of a single base station, the present invention constructs a multi-dimensional joint positioning mathematical model for multi-base station signal fusion positioning based on the digital twin technology of channel multi-domain features. The multi-domain difference between the simulated waveforms of multiple base stations and the actual received waveforms is used as the optimization variable, and the position of the node to be located in the mobile digital world is simulated and tested, and the global optimal result is found through multiple iterations to achieve the position estimation of the node to be located in the physical world.

[0017] 3. The present invention designs a maximum a posteriori probability position estimation method based on dynamic optimization, which can flexibly adjust the optimization strategy according to the hardware computing resources. Under the low computing performance of the positioning system, the present invention uses the negative logarithm of the posterior estimation as the loss function, derives the gradient and Hessian matrix of the loss function in the iterative process, and finally realizes the rapid iteration of the position estimation through the Newton method; under the high computing performance of the positioning system, the positioning system sets the search range and step size, divides the grid, traverses each grid point to calculate the posterior probability, and realizes the optimization of the posterior probability. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a framework diagram for realizing a multi-dimensional joint closed-loop optimization positioning method based on a digital twin of channel multi-domain features according to an embodiment of the present invention;

[0019] Figure 2 1 is a graph showing the experimental results of two optimization methods for the mathematical model of an embodiment of the present invention. DETAILED DESCRIPTION

[0020] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments.

[0021] Aiming at the problems of multipath propagation and non-line-of-sight propagation in complex environments, the present invention proposes a multi-dimensional joint closed-loop optimization positioning method based on channel multi-domain feature digital twins, as shown in the attached figure. Figure 2 As shown. By analyzing the components of multipath signals and considering the delay, attenuation and phase change of each path, the formula for multipath received signals is given. Wavelet transform is used to decompose the received signal in multiple domains, extract features from the time domain, frequency domain and Doppler domain, and define a waveform difference calculation method based on multi-domain features. A mathematical model for multi-base station signal fusion positioning based on multi-domain feature digital twins of channels is constructed, and multi-dimensional joint closed-loop positioning is achieved by deploying multiple base stations in the physical world. The comprehensive difference of multiple base stations is combined with Bayesian inference to derive the posterior probability. In the optimization of the posterior probability, a dynamic optimization mechanism is designed. Based on the hardware resource selection algorithm, the Newton method is derived for fast iterative optimization in a low-performance environment, and the grid search method is used to achieve optimization in a high-performance environment.

[0022] Step 1: Calculation of wavelet transform difference based on multi-domain feature decomposition

[0023] In a complex propagation environment, the propagation path of the signal from the node to be located to the base station may include a direct path as well as multiple reflection and scattering paths. These multipath propagations cause complex changes in the characteristics of the signal at the receiving end. Each path will cause signal propagation delay, intensity attenuation and phase change, so the received signal is the superposition of multiple path signals. For the i-th base station R i , the received signal can be expressed as:

[0024]

[0025] in, is the base station R i The signal received at time t, s(t) is the transmitted signal, N is the number of multiple paths, a i is the attenuation factor of each path, indicating the change in signal strength on the path, θ i is the phase information, ω i is the Doppler rate component, used to capture the target's velocity information, τ i is the time delay of each path signal from the node to be located to the base station through the path. Ray tracing simulation calculates the signal characteristics of different paths to generate simulation waveforms.

[0026] In the positioning system, the node to be located in the physical world T moves to the base stations R1, R2, R n The signal sent and the signal received by the base station are recorded as In the twin world, there is a virtual node T to be located. * and n virtual base stations The received signal is By comparing the waveform signals of the corresponding base stations in the physical world and the twin world and Calculate the difference between the two and use it as feedback information to gradually update the position T of the virtual node to be located * , so that it keeps approaching the position T of the real node to be located.

[0027] By decomposing the signal in the time domain, frequency domain and Doppler domain, the local characteristics and global trends of the signal can be captured at the same time. The present invention uses wavelet transform to analyze multiple base station signals, performs wavelet transform on the waveform signals received in the physical world and the virtual world respectively, and extracts the time domain, frequency domain and Doppler domain coefficients of each base station:

[0028]

[0029] Among them, ψ(t) is the mother wavelet, that is, the basic function of the wavelet, which is used to convolve with the signal to extract features; s is the scale function, which is the inverse of the frequency. The larger the scale, the lower the frequency component, and the smaller the scale, the higher the frequency component; τ is the time parameter, which controls the position of the signal in time; ω is the Doppler coefficient, which reflects the frequency shift effect of the signal; ψ * (t) is the complex conjugate of the mother wavelet. The wavelet transform generates coefficients containing multi-domain information by convolving with the signal φ(t) at different time positions and frequencies.

[0030] Calculate the multi-domain difference ε of the signal waveform received in the physical world and the virtual world based on wavelet transform i :

[0031]

[0032] in, It means summing the differences in different scales, time and Doppler domains. This formula indicates the multi-domain difference between the received signal and the virtual base station signal at each base station, and the sum of the squares of the differences in wavelet coefficients is calculated. Difference ε i (T * ) is smaller, the closer the position of the virtual node to be located is to the real node to be located.

[0033] Step 2: Construction of a mathematical model for multi-base station signal fusion positioning based on channel multi-domain feature digital twins

[0034] The present invention constructs a mathematical model for the multi-dimensional joint closed-loop optimization positioning method of the digital twin of the multi-domain characteristics of the channel. Compared with the existing single-base station closed-loop positioning method, the present invention constructs a multi-dimensional joint closed-loop positioning model by deploying multiple base stations in the physical world.

[0035] Assuming the waveform difference ε i (T * ) obeys Gaussian distribution, and its probability density has an exponential decay form, that is, the smaller the difference, the higher the matching degree, and the greater the corresponding probability; conversely, the larger the difference, the lower the matching degree, and the smaller the corresponding probability. Based on this assumption, a likelihood function based on the difference of multiple waveforms is constructed: Indicates the virtual node position T to be located * Given the situation, the signal waveform received by the i-th base station in the physical world is The probability distribution actually reflects the matching degree between the virtual signal waveform and the real signal waveform. Therefore, the present invention reflects the matching degree through the difference between the virtual signal waveform and the real signal waveform, and constructs the following likelihood function:

[0036]

[0037] in, is the standard deviation of the model error, Represents the waveform received by the i-th base station in the twin world.

[0038] In order to integrate the waveform differences of all base stations, the differences of multiple base stations are jointly modeled. The difference of each base station can be regarded as an independent distribution. Therefore, the joint likelihood function is expressed as the product of the likelihood functions of a single base station. The joint likelihood function is constructed:

[0039]

[0040] Where N is the number of base stations, Φ v (T * ) and Φ p They represent the twin world waveform signals and physical world waveform signals of all base stations respectively. The joint likelihood function integrates the information of all base stations and represents the impact of all base station signals on the positioning of the virtual node to be located. By maximizing the joint likelihood function, the information of multiple base stations can be integrated to obtain the optimal position of the virtual node to be located.

[0041] In addition, to reflect the uncertainty of the node location, the prior probability distribution p(T * ) is approximated as a Gaussian distribution. The Gaussian distribution is not only convenient for subsequent calculations, but also ensures that the prior distribution is sufficiently broad by appropriately selecting a larger covariance, avoiding the possibility that a too narrow distribution limits the true location.

[0042]

[0043] Among them, μ T is the prior mean based on the initial estimated position, V T is the prior covariance matrix.

[0044] Based on the Bayesian criterion, the present invention further constructs the posterior probability distribution of the position of the node to be located, combines the prior information with the joint likelihood function, and provides a theoretical basis for the optimization of the position of the virtual node to be located.

[0045] p(T * |Φ p )∝p(Φ v (T * )|Φ p )·p(T * )(8)

[0046] Step 3: Maximum a posteriori probability position estimation method based on dynamic optimization

[0047] In order to estimate the optimal position of the virtual node to be located, the present invention proposes a posterior probability position estimation method based on dynamic optimization. The method maximizes the posterior probability distribution p(T * |Φ p ) to obtain the optimal position of the virtual node to be located This position is selected as the final positioning result of the node to be positioned, and the target formula is:

[0048]

[0049] To meet the needs of hardware environments with different computing capabilities, the positioning system can automatically select the appropriate optimization algorithm. When the positioning system runs in a low-performance computing environment, the system will switch to the Newton method, which will update the gradient efficiently and iteratively to gradually adjust the position of the node to be located; when the system runs in a high-performance computing environment, the system will switch to the grid search method, which uses a wide search range for global optimization.

[0050] (2) Maximum a posteriori probability optimization mechanism based on Newton’s method

[0051] First, we start from maximizing the posterior probability of the node to be located. To simplify the calculation, we transform it into a loss function that minimizes the negative logarithm form, thereby achieving the goal of maximizing the posterior probability. The loss function is expressed as:

[0052]

[0053] The prior terms are:

[0054]

[0055] The likelihood function of the difference is:

[0056]

[0057] Therefore, the loss function can be expressed as:

[0058]

[0059] In the loss function, the first half is the prior term, which reflects the deviation between the position estimate and the prior mean; the second half is the likelihood function term, which reflects the difference between the base station signals.

[0060] In order to iteratively optimize the position of the virtual node to be located, the Newton method based on the Hassian matrix is ​​used in the step-by-step adjustment process. In order to minimize the loss function, the gradient of the loss function and the Hassian matrix are first calculated.

[0061] gradient Reflects the loss function with respect to position T *The direction of change provides the adjustment of the position of the node to be located T * To minimize the loss function. The gradient of the prior term represents the position of the node to be located T * With its prior mean The gradient is the deviation multiplied by the inverse of the prior covariance matrix:

[0062]

[0063] The gradient of the second half is the waveform difference ε i The derivation of:

[0064]

[0065] Get the gradient of the loss function:

[0066]

[0067] The Hassian matrix H(T * ) is the second-order derivative matrix of the loss function. It reflects the curvature information of the loss function near the position T*. By using the Hassian matrix, the position of the virtual node to be located can be optimized more quickly and accurately. By taking the derivative of each term in the gradient formula again, the outer product term of the gradient and the second-order derivative term of the waveform difference are obtained. The Hassian matrix is ​​obtained:

[0068]

[0069] In the process of optimizing the position of the virtual node to be located, the Newton method is used to quickly iterate to the optimal solution. Represents the loss function L relative to position T * The rate of change of is the first-order derivative of the loss function, which indicates the direction in which the loss function decreases fastest; and the Hassian matrix H(T * ), as the second-order derivative of the loss function, not only provides information about the curvature of the position, but also indicates how to adjust the step size in the direction indicated by the gradient to achieve the best result. In short, the gradient is responsible for guiding the direction, and the Hassian matrix is ​​responsible for adjusting the step size. The two work together to make each step update more accurate and efficient.

[0070] Due to the waveform differences of multiple base stations, i (T * ) is a nonlinear function of the position T*. Direct optimization may lead to slow convergence or unstable optimization. Therefore, we The loss function is expanded by a second order Taylor and approximated as a quadratic function. The second order Taylor expansion formula is:

[0071]

[0072] Through Taylor expansion, the original nonlinear loss function L(T * ) at the current position T * (K) is approximated as a quadratic function, thereby converting the complex nonlinear optimization problem into a simple quadratic optimization problem. After the second-order Taylor expansion, the Newton method is used to iteratively update the position of the virtual node to be located. Newton method update formula:

[0073]

[0074] To ensure that the algorithm stops when convergence is reached, set reasonable termination conditions to prevent unnecessary calculations:

[0075] Termination condition of position change: If the change of the position update of the node to be located is small enough, it is considered to have converged:

[0076]

[0077] Here, ξ is a preset very small threshold, indicating that the position of the virtual node to be located hardly changes;

[0078] Termination condition of loss function change: If the change of loss function is small, it can also be considered to have converged:

[0079]

[0080] Here, δ is a preset threshold, which means that the change in the loss function is already small and further optimization will have little effect;

[0081] Maximum number of iterations termination condition: To prevent the algorithm from falling into an infinite loop, if the number of iterations reaches the maximum allowed value, it is forced to stop.

[0082] By searching for the best value through Newton's method, we can obtain the maximum a posteriori probability, that is, the optimal position estimate:

[0083]

[0084] (2) Maximum a posteriori probability optimization mechanism based on grid search

[0085] In a high-performance computing environment, due to its strong computing power, the positioning system will automatically switch to the grid search method for position optimization. The core idea of ​​the grid search method is to divide a grid evenly in the entire search space, then traverse each grid point, calculate the posterior probability of the corresponding position, and finally select the point with the largest posterior probability as the optimal position estimate.

[0086] In the grid search method, the search space T of the node to be located is first defined as [T x ,T y ,T z ], the search range of each dimension [T i min ,T i max ] and step length ΔT i Defines the resolution of the grid. Suppose you want to go from arrive Search and set the step size to ΔT x , then the candidate position T in the grid n Will be in T x Direction value:

[0087] T x,n =T x,min +n·ΔT x n=0,1,...,N x (twenty three)

[0088] Where N x is the number of grids on the x-axis, similarly T y and T z There are also corresponding grids in the direction.

[0089] By meshing in each dimension, the search space is discretized into a network, and each point in the search space It is a candidate position. The accuracy of the network can be adjusted by the step size. The higher the accuracy, the greater the amount of calculation. Calculate the corresponding posterior probability p(T * |Φ p ), and select the position with the largest posterior probability as the final optimal position estimate.

[0090]

[0091] Through this dynamic optimization method of posterior probability position estimation, the position estimation of the node to be located can flexibly adapt to the existing hardware resources and achieve performance optimization and balance in different environments.

[0092] In order to verify the performance of the mathematical model of the present invention, the actual position of the node T to be located is [5, 1, 0], which is assumed to be unknown. The positions of the three base stations R1, R2 and R3 are [5, 7, 1], [9, -1, 1] and [0, -1, 1] respectively, and a digital space for electromagnetic signal propagation is constructed. It is assumed that the initial position of the virtual node to be located is [0, 0, 0].

[0093] In the experiment, the CloudRT platform was used to simulate the signal transmission between the node to be located and the base station. The simulation used the ultra-wideband (UWB) band signal as the transmission signal, set the center frequency of the transmission signal to 6GHz, the signal bandwidth to 500MHz, and the sampling rate to 1GS / s. In the simulation scenario, the obstacle materials include metal, wood, and glass, representing high reflectivity, partial transmittance, and high transmittance characteristics, respectively. The simulation generated the signal received by each base station, including the delay, attenuation, and phase change of the direct path and the multiple reflection path.

[0094] The experimental results of the two optimization methods are shown in the attached Figure 2 As shown. When using the gradient descent method for optimization, the system adjusts the position according to the prior information and the gradient of the waveform difference. Through multiple iterations, the position of the virtual node to be located gradually converges, and the final position is close to the position of the real node to be located. After optimization, the position of the virtual node to be located is [5.07, 1.01, -0.00], and the positioning error is about 0.071 meters. When switching to the grid search method for global optimization, by evaluating the posterior probability of each position, the position of the virtual node to be located is [4.98, 1.03, 0.00], and the positioning error is reduced to about 0.036 meters.

Claims

1. A multi-dimensional joint closed-loop optimization positioning method based on channel multi-domain feature digital twins, characterized in that: The steps include: Step 1: Calculation of wavelet transform difference based on multi-domain feature decomposition By analyzing the components of multipath signals and considering the delay, attenuation and phase change of each path, the formula for multipath received signals is given; by constructing a channel digital twin system, the signal characteristics of different paths are extracted to generate simulation waveforms, and the received physical and virtual signals are decomposed into multiple domains using wavelet transform to extract the delay characteristics of the time domain, the attenuation characteristics of the frequency domain and the phase change characteristics of the Doppler domain, and the multi-domain wavelet coefficients are obtained; the difference of the wavelet coefficients is used as the multi-domain difference of the signal to construct a likelihood function based on the multi-domain difference; Step 2: Construction of a multi-dimensional joint positioning mathematical model based on channel multi-domain feature digital twins A mathematical model is constructed for the multi-dimensional joint closed-loop optimization positioning method of the digital twin of the multi-domain characteristics of the channel. Compared with the single-base station closed-loop positioning method that has been studied, a multi-dimensional joint closed-loop optimization positioning model is constructed by deploying multiple base stations in the physical world and mapping them to the virtual world; Gaussian distribution is used to represent the likelihood function of the multi-domain difference, and considering the joint effect of multiple base stations, a multi-dimensional joint likelihood function based on the multi-domain difference is formed; the prior estimate of the node to be located is set, and the Bayesian inference method is used to derive the posterior probability based on the channel digital twin, and a mathematical model based on multi-base station signal fusion positioning is constructed; Step 3: Maximum a posteriori probability position estimation method based on dynamic optimization In order to meet the needs of hardware environments with different computing capabilities, a dynamic optimization method for maximum a posteriori probability position estimation is designed. When the positioning system runs in a low-performance computing environment, the negative logarithm of the posterior estimate is used as the loss function, the gradient and Hessian matrix of the loss function are derived during the iteration, and the fast iteration of position estimation is achieved through the Newton method. When the system runs in a high-performance computing environment, the search range and step size are set, and the maximum a posteriori probability position estimation is achieved by dividing the grid and traversing each grid point to calculate the posterior probability. By dynamically selecting the optimization algorithm, the positioning system can automatically adjust the algorithm according to hardware resources, taking into account both computing efficiency and positioning accuracy.

2. The method according to claim 1, characterized in that In the step 1, wavelet transform is used to analyze multiple base station signals, and wavelet transform is performed on the waveform signals received in the physical world and the virtual world respectively to extract the time domain, frequency domain and Doppler domain coefficients of each base station: in, It is the physical world base station R i The signal received at time t is It is the virtual world base station R i The signal received at time t, ψ(t) is the mother wavelet, that is, the basic function of the wavelet, which is used to convolve with the signal to extract features; s is the scale function, which is the inverse of the frequency. The larger the scale, the lower the frequency component, and the smaller the scale, the higher the frequency component; τ is the time parameter, which controls the position of the signal in time; ω is the Doppler coefficient, which reflects the frequency shift effect of the signal; ψ * (t) is the complex conjugate of the mother wavelet. The wavelet transform generates coefficients containing multi-domain information by convolving with the signal φ(t) at different time positions and frequencies. Calculate the multi-domain difference ε of the signal waveform received in the physical world and the virtual world based on wavelet transform i : in, It represents the summation of differences in different scales, time and Doppler domains.

3. The method according to claim 1, characterized in that In the step 2, Gaussian distribution is used to represent the likelihood function of multi-domain difference Indicates the virtual node position T to be located * Given the situation, the signal waveform received by the i-th base station in the physical world is The probability distribution of is as follows: in, is the standard deviation of the model error, Represents the waveform received by the i-th base station in the twin world. In order to integrate the waveform differences of all base stations, the differences of multiple base stations are jointly modeled. The difference of each base station can be regarded as an independent distribution. Therefore, the joint likelihood function is expressed as the product of the likelihood functions of a single base station. The joint likelihood function is constructed: Where N is the number of base stations, Φ v (T * ) and Φ p They respectively represent the twin world waveform signals and physical world waveform signals of all base stations. To reflect the uncertainty of the node location, the prior probability distribution p(T * ) is approximately Gaussian distribution, and the formula is as follows: Among them, μ T is the prior mean based on the initial estimated position, V T is the prior covariance matrix. Based on the Bayesian criterion, the present invention further constructs the posterior probability distribution of the position of the node to be located, combines the prior information with the joint likelihood function, and provides a theoretical basis for the optimization of the position of the virtual node to be located. p(T * |F p )∝p(Φ v (T * )|F p )·p(T * )。 4. The method according to claim 1, characterized in that In the step 3, based on the maximum a posteriori probability optimization mechanism of the Newton method, a loss function in the form of negative logarithm of the posterior probability of the position of the node to be located is taken, and the loss function is expressed as: Priors: The likelihood function of the difference is: Therefore, the loss function can be expressed as: In the loss function, the first half is the prior term, which reflects the deviation between the position estimate and the prior mean; the second half is the likelihood function term, which reflects the difference between the base station signals. gradient Reflects the loss function with respect to position T * The direction of change, the gradient of the prior term represents the position of the node to be located T * With its prior mean μ T* The gradient is the deviation multiplied by the inverse of the prior covariance matrix: The gradient of the second half is the waveform difference ε i The derivation of: Get the gradient of the loss function: The Hassian matrix H(T * ) is the second-order derivative matrix of the loss function. It reflects the curvature information of the loss function near the position T*. Newton's method update formula: In order to ensure that the algorithm stops when it reaches convergence, a reasonable termination condition is set. Finally, the maximum a posteriori probability, that is, the optimal position estimate, is obtained by optimizing through Newton's method:

5. The method according to claim 1, characterized in that In step 3, based on the maximum a posteriori probability optimization mechanism of the grid search method, the search space T of the node to be located is first defined as [T x ,T y ,T z ], the search range for each dimension and step length ΔT i Defines the resolution of the grid. Suppose you want to go from arrive Search and set the step size to ΔT x , then the candidate position T in the grid n Will be in T x Direction value: T x,n =T x,min +n·ΔT x n=0,1,...,N x Where N x is the number of grids on the x-axis, similarly T y and T z There are also corresponding grids in the direction. By meshing in each dimension, the search space is discretized into a network, and each point in the search space It is a candidate position. The accuracy of the network can be adjusted by the step size. The higher the accuracy, the greater the amount of calculation. Calculate the corresponding posterior probability p(T * |Φ p ), and select the position with the largest posterior probability as the final optimal position estimate. Through this dynamic optimization method of posterior probability position estimation, the position estimation of the node to be located can flexibly adapt to the existing hardware resources and achieve performance optimization and balance in different environments.

Images