MLE and UKF combined speed perception and prediction beam forming method oriented to Internet of Vehicles

Abstract

The invention discloses an Internet of Vehicles-oriented MLE and UKF combined speed sensing and prediction beam forming method, which comprises the following steps of: establishing a two-dimensional coordinate model of a uniform linear array in a near-field uplink narrow-band transmission scene, and accurately describing a relative position and a motion state between a target and a base station; according to a near-field channel model, calculating propagation characteristics of echo signals received by each antenna unit; a speed estimation problem is converted into an unconstrained optimization problem to be solved by taking minimization of a loss function of observation data as a target, and the speed of a target vehicle is estimated by using received echo signals and target expected position information; taking the speed estimated by the MLE from the echo signal as an observed value of the UKF, and iteratively optimizing the estimation precision of the state of the target vehicle in combination with a state prediction and updating mechanism of the UKF; and according to the optimized motion state of the target vehicle, a beam forming scheme is designed, data transmission is carried out, and communication rates under different conditions are evaluated. According to the invention, prediction precision and communication performance in a mobile scene are effectively improved.

Description

Technical Field

[0001] This invention relates to a method for joint speed sensing and predictive beamforming using MLE and UKF for vehicle-to-everything (V2X) communication, belonging to the field of wireless communication technology. Background Technology

[0002] Near-field velocity sensing is a technique in wireless communication systems that utilizes near-field propagation characteristics to estimate the velocity of a moving target. Compared to far-field sensing, near-field sensing can simultaneously estimate the radial and lateral velocities of a moving target. Furthermore, in the near-field region, spherical wave propagation introduces a new range dimension, potentially facilitating joint estimation of range and angle in sensing. Target velocity estimation depends on the estimation of the Doppler frequency. Compared to far-field sensing, near-field sensing observes the target from different directions. Two antennas in a large-scale antenna array that are far apart may have significantly different Doppler frequencies, thus enhancing the estimation of the moving velocity.

[0003] In high-mobility communication systems, target tracking is an effective method that ensures stable communication and sensing performance while meeting low latency requirements. Sensing-assisted predictive beamforming can maintain alignment without a dedicated backhaul link, eliminating the need for pilot signals. Predictive beamforming can predict a user's motion state in subsequent coherent time intervals based on the user's state in the current coherent time interval (CPI). The user's current motion state can be obtained from the echo signal. By utilizing predictive beamforming technology, the timeliness of beamformers can be improved without a dedicated feedback link.

[0004] A literature search of existing technologies revealed that Zhen Du et al., in their paper titled "Integrated Sensing and Communications for V2I Networks: Dynamic Predictive Beamforming for Extended Vehicle Targets" published in *IEEE Transactions on Wireless Communications, vol. 22, no. 6, June 2023*, used the Extended Kalman Filter (EKF) method to predict vehicle position. However, this paper considers far-field scenarios and can only estimate the vehicle's radial velocity; the lateral velocity is usually not discernible from the echo signal. Further research revealed that Zhaolin Wang et al., in their paper titled "Near-Field Velocity Sensing and Predictive Beamforming" published in *IEEE Transactions on Vehicular Technology, vol. 74, no. 1, Jan. 2025*, proposed a maximum likelihood-based method for jointly estimating radial and lateral velocities from echo signals. However, using MLE alone for velocity estimation can be affected by noise and short-term dynamic changes, leading to unstable estimations or distortions at certain times. Therefore, this invention combines MLE and UKF for velocity sensing and predictive beamforming, dynamically updating the system state. By combining the predictive model and observation data, UKF can smoothly track the target's motion and reduce estimation fluctuations and errors. Therefore, by combining UKF, the system can estimate the target's velocity more accurately and overcome the accuracy degradation that may occur when using MLE alone. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings and deficiencies of the existing technology by proposing a joint speed sensing and prediction beamforming method based on MLE and UKF for vehicle networking. This method is applied to communication transmission between automated guided vehicles (AGVs) in factories and fixed base stations. It provides a high-precision, low-overhead, pilot-signal-free communication scheme and can be used to improve the accuracy of target speed estimation and beamforming in near-field communication systems.

[0006] The technical solution adopted by this invention to solve its technical problem is: a method for joint speed sensing and prediction beamforming of MLE and UKF for vehicle networking, which includes the following steps:

[0007] Step 1: In the near-field uplink narrowband transmission scenario, establish a two-dimensional coordinate model of a uniform linear array to accurately describe the relative position and motion state between the target and the base station;

[0008] Step 2: Calculate the propagation characteristics of the echo signal received by each antenna element based on the near-field channel model;

[0009] Step 3: With the goal of minimizing the loss function of the observed data, the speed estimation problem is transformed into an unconstrained optimization problem. The speed of the target vehicle is estimated using the received echo signal and the expected position information of the target.

[0010] Step 4: Use the speed estimated from the echo signal by MLE as the observation value of UKF, and combine the state prediction and update mechanism of UKF to iteratively optimize the estimation accuracy of the target vehicle state;

[0011] Step 5: Based on the optimized target vehicle motion state (position and speed), design an appropriate beamforming scheme, perform data transmission, and evaluate the communication rate under different conditions.

[0012] Furthermore, step 1 of the present invention specifically includes:

[0013] In a near-field uplink narrowband transmission scenario, the base station is equipped with a uniform linear antenna array (ULA) with M antennas spaced d apart. The ULA deployed at the base station is spread along the X-axis, with the center of the ULA at the origin of the coordinate system. The signal propagation distance between the m-th antenna at the base station and the target is r. m The velocity component of the target projected by the line connecting the base station and the m-th antenna at the target location is v. m :

[0014]

[0015] The coordinates of the m-th antenna at the base station are (δ m d,0), where v r,m and v θ,m These are the projections of the radial and lateral velocities along the line connecting the m-th antenna and the target, respectively, g m and q m Represents projection coefficient:

[0016]

[0017] v r,m =g m v r ,v θ,m =q m v θ (3)

[0018] v m =vr,m +v θ,m (4)

[0019] Furthermore, step 2 of the present invention specifically includes:

[0020] The baseband echo signal received on the m-th antenna by the moving target is:

[0021]

[0022] Among them, channel gain β is constant within a CPI, s i (n) represents the transmitted signal of the i-th antenna, z m (n) represents Gaussian noise.

[0023] The overall received echo signal is y(n) = [y1(n),...,y M (n)] T :

[0024] y(n)=β(A(r,θ)⊙D n (v r ,v θ ))s(n)+z(n) (6)

[0025] Where, A(r,θ)=a(r,θ)a T (r,θ), where a(r,θ) represents the near-field array response. d n (v r ,v θ ) represents the Doppler frequency shift vector, and z(n) represents the noise vector.

[0026]

[0027] Further, step 3 of the present invention specifically includes:

[0028] To maximize the likelihood function, we need to minimize the sum of squares of the error Y - βX(η,v):

[0029]

[0030] Among them, ||·|| F Let represent the Frobenius function. For a given velocity v, first calculate the optimal estimate of β, obtaining:

[0031]

[0032] By solving the above problem, we can obtain The optimal solution:

[0033]

[0034] Then, Substituting into the loss function, we obtain the optimization problem for v:

[0035]

[0036] Further, step 4 of the present invention specifically includes:

[0037] Assume the state of the AGV is x k =[r k ,θ k ,v r,k ,v θ,k ] T That is, the target's position and velocity. The state transition equation describes the state change from time k to k+1:

[0038] x k+1|k =f(x) k (13)

[0039] The specific state transition equations are as follows:

[0040] r k+1 =r k +v r,k Δt (14)

[0041]

[0042] v r,k+1 =v r,k (16)

[0043] v θ,k+1 =v θ,k (17)

[0044] Set initial state estimation Given the covariance matrix P0, use the UT transformation to generate a sigma point set and calculate the state prediction values ​​for these points; first, calculate the parameter λ, λ = α, which controls the range of the sigma point expansion. 2 (L+κ)-L, where α is the extended range of the control points, L is the dimension of the state vector, and κ is an adjustable parameter. Next, the extended matrix S is calculated to generate sigma points. The formula for generating sigma points is: The formula for calculating the weight is: Among them, W m It is the mean weight, W c It is the covariance weight, used to calculate the weighted average of sigma points.

[0045] For each sigma point Through the state transition function f(x) k Perform state propagation:

[0046]

[0047] Use the weight W of the sigma point m The weighted average of the predicted states for each sigma point is used to obtain the mean of the state predictions: The predicted covariance matrix is ​​obtained by calculating the weighted covariance at each sigma point:

[0048]

[0049] Where Q is the process noise covariance matrix.

[0050] The observables in the observation space are velocity estimates (obtained through MLE), therefore the observation function is: By observing the function h(x) k Mapping the sigma point from the state space to the observation space:

[0051]

[0052] Use the weight W of the sigma point m The predicted mean of the observations is obtained by taking a weighted average of the observations:

[0053]

[0054] The observation covariance matrix is ​​calculated by adding the observation noise covariance matrix R to the difference between the predicted observations and the observed mean using sigma.

[0055]

[0056] Cross covariance matrix P xz It is used to measure the relationship between the state and the observation, and it is obtained by calculating the difference between the state and the observation for each predicted sigma point:

[0057]

[0058] The formula for calculating Kalman gain is: Update the state estimate based on the observations and Kalman gain:

[0059]

[0060] The updated covariance matrix is:

[0061]

[0062] Further, step 5 of the present invention specifically includes:

[0063] By adjusting the weights of the antenna array, the transmitted signal can be concentrated with energy and directed towards the target's direction of motion, thereby maximizing signal transmission efficiency. Beamforming weights can be calculated as follows:

[0064]

[0065] Wherein, D(v) r ,v θ ) is determined by the target velocity (v) r ,v θ The calculated Doppler frequency shift matrix, a H (r,θ) is the transpose complex conjugate array response of the antenna array.

[0066] Communication rate can be calculated using the following formula:

[0067]

[0068] Where N is the number of symbol periods, h H (n)=βa T (r,θ)diag(v r ,v θ ) is the communication channel, and w(n) is the beam weight.

[0069] Beneficial effects:

[0070] 1. This invention provides a high-precision, low-overhead, pilot-signal-free communication scheme that can be applied to improve the accuracy of target velocity estimation and beamforming in near-field communication systems.

[0071] 2. This invention realizes an efficient nonlinear state estimation and beamforming framework, which effectively improves prediction accuracy and communication performance in mobile scenarios. Attached Figure Description

[0072] Figure 1 This is a flowchart of the speed sensing method combining MLE and UKF of the present invention.

[0073] Figure 2 This is a communication system model diagram of the method of the present invention.

[0074] Figure 3 This is a schematic diagram of the simulation scene coordinates for the method of the present invention.

[0075] Figure 4 This is a comparison chart of the radial velocity predicted by the method of the present invention and the actual radial velocity.

[0076] Figure 5This is a comparison chart of the lateral velocity predicted by the method of this invention and the actual lateral velocity.

[0077] Figure 6 This is a comparison diagram of the trajectory predicted by the method of the present invention and the actual trajectory.

[0078] Figure 7 This is a comparison chart showing the achievable rates of the method of the present invention with and without Doppler frequency compensation (DFC). Detailed Implementation

[0079] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0080] Step 1: In a near-field uplink narrowband transmission scenario, the base station is equipped with a uniform linear antenna array, which consists of 256 antenna elements. The antenna element spacing d is set to half a wavelength, and the system center frequency f is... c The system bandwidth B is 500kHz, and the symbol duration is 28GHz. The durations of N = 1000 symbols are grouped into a single CPI, with each CPI having a length of ΔT = NT. s =2ms.

[0081] A polar coordinate model is used to characterize the channel model between the antenna array elements and the AGV. The ULA is expanded along the X-axis, with the origin at the center of the ULA. The coordinates of the m-th antenna at the base station are (δ... m d,0), where The AGV's position is (r, θ), where r represents the distance of the target relative to the center of the ULA, and θ represents the angle of the target relative to the center of the ULA.

[0082] The signal propagation distance between the m-th antenna at the base station and the target is r. m The velocity component of the target projected by the line connecting the base station and the m-th antenna at the target location is v. m :

[0083]

[0084] Step 2: The baseband echo signal received on the m-th antenna reflected by the moving target is:

[0085]

[0086] Among them, channel gain β is constant within a CPI, satisfying the condition that... Transmit antenna gain G t =1, receiving antenna gain G r =1, radar cross-section σ RCS= -23dB,s i (n) represents the transmitted signal of the i-th antenna, z m (n) represents Gaussian noise, and the noise density power is set to -174dBm / Hz.

[0087] The overall received echo signal is y(n) = [y1(n),...,y M (n)] T :

[0088] y(n)=β(A(r,θ)⊙D n (v r ,v θ ))s(n)+z(n) (3)

[0089] Where, A(r,θ)=a(r,θ)a T (r,θ), where a(r,θ) represents the near-field array response. d n (v r ,v θ ) represents the Doppler frequency shift vector, and z(n) represents the noise vector.

[0090]

[0091] Step 3: To maximize the likelihood function, we need to minimize the sum of squares of the error Y - βX(η,v):

[0092]

[0093] Among them, ||·|| F Let represent the Frobenius function. For a given velocity v, first calculate the optimal estimate of β, obtaining:

[0094]

[0095] By solving the above problem, we can obtain The optimal solution:

[0096]

[0097] Then, Substituting into the loss function, we obtain the optimization problem for v:

[0098]

[0099] MLE optimizes the function To estimate the velocity, gradient descent is used to update the velocity estimate during optimization. The update formula for gradient ascent is:

[0100]

[0101] Where α(t) is the learning rate, It is the gradient of the objective function g with respect to the velocity v.

[0102] Step 4: Assume the state of the AGV is x k =[r k ,θ k ,v r,k ,v θ,k ] T That is, the target's position and velocity. The state transition equation describes the state change from time k to k+1:

[0103] x k+1|k =f(x) k (11)

[0104] The specific state transition equations are as follows:

[0105] r k+1 =r k +v r,k Δt (12)

[0106]

[0107] v r,k+1 =v r,k (14)

[0108] v θ,k+1 =v θ,k (15)

[0109] Set initial state estimation Covariance Matrix The UT transformation is used to generate a set of sigma points, and the state predictions of these points are calculated. First, the parameter λ, λ = α, which controls the range of the sigma point expansion is calculated. 2 (L+κ)-L, where α is the extended range of the control point, taking a value of 1e. -3 L is the dimension of the state vector, and κ is an adjustable parameter, set to 0. Next, the extended matrix S is calculated to generate sigma points. The formula for generating sigma points is: The formula for calculating the weight is: Among them, W m It is the mean weight, W c It is the covariance weight, used to calculate the weighted average of sigma points.

[0110] For each sigma point Through the state transition function f(x) k Perform state propagation:

[0111]

[0112] Use the weight W of the sigma point m The weighted average of the predicted states for each sigma point is used to obtain the mean of the state predictions: The predicted covariance matrix is ​​obtained by calculating the weighted covariance at each sigma point:

[0113]

[0114] in, It is the process noise covariance matrix.

[0115] The observables in the observation space are velocity estimates (obtained through MLE), therefore the observation function is: By observing the function h(x) k Mapping the sigma point from the state space to the observation space:

[0116]

[0117] Use the weight W of the sigma point m The predicted mean of the observations is obtained by taking a weighted average of the observations:

[0118]

[0119] The observation covariance matrix is ​​calculated by adding the observation noise covariance matrix to the difference between the predicted observations and the observed mean, expressed as sigma.

[0120]

[0121] Cross covariance matrix P xz It is used to measure the relationship between the state and the observation, and it is obtained by calculating the difference between the state and the observation for each predicted sigma point:

[0122]

[0123] The formula for calculating Kalman gain is: Update the state estimate based on the observations and Kalman gain:

[0124]

[0125] The updated covariance matrix is:

[0126]

[0127] Step 5: For predicted beamforming, we can calculate the beam pointing based on the target's motion state. Beamforming weights can be calculated as follows:

[0128]

[0129] Wherein, D(v) r ,v θ ) is determined by the target velocity (v) r ,v θ The calculated Doppler frequency shift matrix, a H (r,θ) is the transpose complex conjugate array response of the antenna array.

[0130] To predict target motion and optimize beam pointing, the target's velocity affects the beamforming design. By estimating the velocity and predicting the position from the previous moment, the base station can design beam weights that adapt to the target's movement.

[0131]

[0132] Optimal beamforming weights are calculated based on the target's position and velocity, aiming to minimize the Doppler effect and maximize signal focusing. The formula for calculating the optimal beam is:

[0133]

[0134] Communication rate can be calculated using the following formula:

[0135]

[0136] Where N = 1000 is the number of symbol periods, h H (n)=βa T (r,θ)diag(v r ,v θ ) is the communication channel, and w(n) is the beam weight.

[0137] This invention places the AGV 10 meters from the center of the base station, with an azimuth angle of... In this section, the system simulates signal transmission and reception by performing motion modeling on a single AGV user and considering multiple antenna arrays and correlation time. The system model diagram is shown below. Figure 2 As shown, the base station transmits signals to the AGV via beamforming and receives the echo, while simultaneously predicting the AGV's next state. The AGV's movement path is simulated using a pre-defined circular trajectory, and its speed is determined by the actual speed and the radius of the circular trajectory. The base station employs a full-duplex antenna architecture, where a circulator is used to achieve simultaneous transmission and reception. Figure 3 As shown.

[0138] In this invention, the initial state is (x,y)=(8,6) and the initial speed is 20m / s. The radial and lateral speeds of the AGV are estimated based on UKF and these speeds are independently estimated using the MLE method. Figure 4 and Figure 5 The estimated radial and lateral velocities of the AGV at each CPI are displayed. Both velocities are very close to the ground truth, and the lateral velocity sensing is more sensitive to the BS distance of the AGV compared to the radial sensing. Figure 6 The image shows a comparison between the trajectory predicted by the MLE algorithm and the trajectory predicted by the combined MLE and UKF algorithms and the actual trajectory. It can be seen that the trajectories predicted by both algorithms are very close to the ground truth, but the trajectory predicted by the combined MLE and UKF algorithms is more accurate than the trajectory predicted by the MLE algorithm alone. Figure 7 The comparison of communication rates of two different algorithms with the optimal communication rate is shown, and the communication performance degrades when lateral velocity dominates without Doppler frequency compensation.

Claims

1. A method for joint speed sensing and prediction beamforming using MLE and UKF for vehicle-to-everything (V2X) communication, characterized in that, Includes the following steps: Step 1: In the near-field uplink narrowband transmission scenario, establish a two-dimensional coordinate model of a uniform linear array to accurately describe the relative position and motion state between the target and the base station; Step 2: Calculate the propagation characteristics of the echo signal received by each antenna element based on the near-field channel model; Step 3: With the goal of minimizing the loss function of the observed data, the speed estimation problem is transformed into an unconstrained optimization problem. The speed of the target vehicle is estimated using the received echo signal and the expected position information of the target. Step 4: Use the speed estimated from the echo signal by MLE as the observation value of UKF, and combine the state prediction and update mechanism of UKF to iteratively optimize the estimation accuracy of the target vehicle state; Step 5: Based on the optimized target vehicle motion state, design an appropriate beamforming scheme, perform data transmission, and evaluate the communication rate under different conditions.

2. The method for joint speed sensing and prediction beamforming of MLE and UKF for vehicle-to-everything (V2X) networks according to claim 1, characterized in that, Step 2 includes: The baseband echo signal received on the m-th antenna by the reflection from the moving target is y. m The received overall echo signal Y(n) can be represented as the transmitted signal S(n), the channel response matrix A(r,θ), and the Doppler frequency shift vector D. n (v r ,v θ The function of the noise vector z(n), specifically, y(n) = β(A(r,θ)⊙D n (v r ,v θ ))s(n)+z(n), where ⊙ represents the Hadamard product.

3. The method for joint speed sensing and prediction beamforming of MLE and UKF for vehicle-to-everything (V2X) networks according to claim 1, characterized in that, Step 3 includes: Based on the echo signal Y(n) and the known base station antenna array configuration, a maximum likelihood estimation problem for the target velocity is constructed, with the objective of minimizing the error between the observed data Y(n) and the model prediction data based on the velocity vector v. For a given velocity v, estimate the channel gain β. Obtain the optimal estimate of β by maximizing the likelihood function. Substitute the obtained optimal estimate into the objective function to obtain an optimization problem that is only related to the velocity vector v.

4. The method for joint speed sensing and prediction beamforming of MLE and UKF for vehicle networking according to claim 1, characterized in that, The method transforms the optimization problem into an unconstrained optimization problem, including: By minimizing the loss function of the observed data, the radial and lateral velocities of the system can be estimated, which can be transformed into solving an unconstrained optimization problem: Where g(Y,η,v) is the maximum likelihood function used to estimate v, Y represents the echo signal received at a CPI, and η = [r,θ] T This indicates the position of the AGV (Automated Guided Vehicle).

5. A method for joint speed sensing and prediction beamforming of MLE and UKF for vehicle-to-everything (V2X) networks according to claim 1, characterized in that, Step 4 includes: Set initial state estimation The covariance matrix P0 describes the target's state at the initial time and the uncertainty of the state estimate; a sigma point set is generated using UT transformation, and the weights of the sigma points are calculated, including the mean weight W. m Covariance weights W c , used for state prediction and updating; Use the weight W of the sigma point m The state predictions for each sigma point are weighted and averaged to obtain the mean state prediction. The predicted covariance matrix is ​​obtained by calculating the weighted covariance at each sigma point.

6. The method for joint speed sensing and prediction beamforming of MLE and UKF for vehicle-to-everything (V2X) networks according to claim 1, characterized in that, The method uses the velocity estimated from the echo signal by the MLE as the UKF observation, including: The observables in the observation space are velocity estimates (obtained through MLE), therefore the observation function is: By observing the function h(x) k Mapping sigma points from the state space to the observation space; Use the weight W of the sigma point m The observed values ​​are weighted and averaged to obtain the predicted observed mean; the observed covariance matrix is ​​calculated by the difference between the predicted observed values ​​and the observed mean using sigma. The cross-covariance matrix P is obtained by calculating the difference between the state and the observed state at each predicted sigma point. xz It is used to measure the relationship between state and observation; Calculate the Kalman gain using the cross covariance matrix and the observation covariance matrix: Update the state estimate and covariance matrix based on the observed values ​​and Kalman gain; The above state prediction and update steps are used as an iterative process of unscented Kalman filtering. For each new observation time, these steps are repeated to continuously and dynamically track the state changes of the target.

7. A method for joint speed sensing and prediction beamforming of MLE and UKF for vehicle-to-everything (V2X) networks according to claim 1, characterized in that, Step 5 includes: In the initial stage, the beamformer is formulated by estimating the target's velocity in the first CPI and based on the initial position information. l (n), set the iteration counter to l = 1; Repeat the following steps: using w l (n) Transmit data to the user and collect echo signal Y l The radial and lateral velocities are estimated, and the user's next position is predicted using MLE and UKF, thus enabling the design of the next beamformer. l+1 (n), update the iteration counter l = l + 1 until the user moves out of the base station's coverage area.

Images