Unmanned aerial vehicle positioning method based on communication base station beam scanning

Abstract

The invention discloses a method for realizing target positioning of an unmanned aerial vehicle based on a communication base station beam scanning mode. In order to solve the security problems and challenges brought to public security, privacy protection, aviation and key infrastructures by black flight of the unmanned aerial vehicle, the system is deployed on the communication base station, and a beam scanning method is adopted to realize positioning of the unmanned aerial vehicle. In the first stage, a base station periodically performs beam scanning in an observation area, and coarse positioning of an unmanned aerial vehicle is realized through echo data processing; and in the second stage, based on a coarse positioning result in the first stage, a motion model is combined, and an optimization and filtering method is utilized to obtain a complete motion track, so that two-dimensional positioning of the unmanned aerial vehicle is realized. The method has the advantages that communication and perception are integrated, the hardware system deployment cost is low, and the algorithm complexity is low.

Description

Technical Field

[0001] The invention belongs to the field of electronic signal processing and relates to a UAV positioning method based on communication base station beam scanning. Background Art

[0002] With the rapid development of information technology, drone technology has expanded from the military field to the civilian field, and its application scope is becoming increasingly wide, covering aerial photography, agriculture, logistics, environmental protection and other fields. Drones have been widely used in many fields such as military and civilian due to their easy operation, strong maneuverability and wide application range. However, with the continuous advancement and widespread application of drone technology, the problem of unauthorized flight (i.e. "black flight") has become increasingly serious, bringing many challenges to society. "Black flight" refers to the illegal flight of drones in no-fly zones, restricted flight zones or other sensitive areas without permission or in violation of regulations. Such behavior poses a serious threat to public safety, privacy protection, aviation safety and critical infrastructure. Therefore, how to effectively monitor drones and prevent their potential risks has become an important issue that needs to be solved urgently.

[0003] At present, the technologies for drone positioning mainly include optics, radio, acoustics, and radar. These technologies are effective in specific scenarios, but they also have obvious limitations: systems based on optical signals can directly obtain visual images of the target, but are greatly affected by the environment and weather and are not suitable for long-distance application scenarios; radio-based systems have a high false alarm rate in complex communication environments; acoustic-based systems are greatly affected by environmental noise and have a short detection distance. Radar-based systems have a long detection distance and high accuracy, but the deployment cost is high.

[0004] The present invention designs a method for positioning the target of an unmanned aerial vehicle by using a communication base station beam scanning method. The method is generally divided into two stages. In the first stage, the base station periodically performs beam scanning in the observation area to obtain the target scattering coefficient matrix, thereby realizing the rough positioning of the base station. In the second stage, based on the target position obtained by the rough positioning, the motion model and position information are combined by particle swarm optimization and extended Kalman filtering methods, and the rough positioning points are optimized to obtain the complete point trajectory after optimization and prediction, thereby realizing the two-dimensional positioning of the unmanned aerial vehicle. Summary of the invention

[0005] The purpose of the present invention is to realize the positioning of the UAV target by using the communication base station beam scanning method. The method is generally divided into two stages. The first stage obtains the rough positioning point of the target by scanning the base station. The second stage realizes two-dimensional positioning of the UAV target according to the target's motion model based on the rough positioning result combined with particle swarm optimization, Kalman filtering and other methods. Specifically, the following steps are included:

[0006] The first stage is the base station's rough positioning of the drone:

[0007] Step 1: Set up the system and its working mode. The system is deployed on the base station in a side-mounted manner. The radar transceiver antennas in the system are placed in the same position, and the radar works in electronic scanning mode. The transmitting platform turns on the beam scanning working mode, switching the beam in the elevation and azimuth directions in a fixed period on a pre-set monitoring plane. The receiving platform uses a fixed beam working mode. The system working concept diagram is as follows: Figure 1 shown.

[0008] Step 2: The transmitting platform transmits the OFDM signal in the millimeter wave frequency band, and the receiving platform receives the echo signal reflected by the target. After removing the carrier frequency, the received echo signal can be expressed as

[0009]

[0010] In formula (1), t is the fast time, τ is the slow time, N is the number of subcarriers, and f c is the carrier frequency, c is the speed of light, λ is the wavelength, Δf is the subcarrier spacing, T is the pulse width, R(τ) is the slant range between the target and the radar, represents the modulation function of the antenna pointing, θ P and Respectively represent the azimuth and elevation angle of the target at that moment. Formula (1) can be expressed as the convolution relationship between the target scattering and the point spread function, which can be expressed in matrix form as

[0011] y=HsZ T (2)

[0012] Then the target scattering matrix is ​​restored, and the solution of the target optimization function can be obtained by the least squares method:

[0013]

[0014] Step 3: Find the peak position of the target scattering coefficient matrix amplitude in equation (3), then the two-dimensional coordinate position of the target is the two-dimensional coordinate position of the target scattering coefficient matrix corresponding to the peak value:

[0015]

[0016] where x i ,y i is the two-dimensional positioning coordinate point of the target in the i-th frame.

[0017] The second stage is the optimization of the drone trajectory based on the base station's rough positioning parameters:

[0018] Step 4: Pre-filter the rough positioning trajectory, smooth the trajectory and remove unreasonable points. Apply the average sliding filter to smooth the entire trajectory, reduce the impact of unreasonable trajectory points on the trajectory, and eliminate sharp changes. Then optimize the trajectory based on the speed threshold method, and use the average value obtained by the local average method to replace unreasonable trajectory points.

[0019] Step 5: Generate a point cloud around the two-dimensional coordinates of each positioning point obtained by rough positioning, and randomly select points from each point cloud group to form a trajectory, such as Figure 2 As shown in the figure, multiple trajectories are selected as the initial particle swarm, and the particle swarm optimization algorithm is used to optimize the coarse positioning trajectory based on the known target motion model and the root mean square error (RMSE) of the polynomial fitting is used as the cost function.

[0020] Step 6: Based on the optimized trajectory results, use the position backward difference method to obtain the two-dimensional average speed of the moving target. The time series corresponding to the average speed and the average speed are used as the training data for Gaussian process regression, and the complete time series is input to predict the speed of the first position. The speed between two positions is considered to be equal to the speed of the previous position.

[0021] Step 7: Extract distance information from the original echo signal. Formula (1) performs FFT along the distance direction to perform distance compression. The compressed signal is in the form of

[0022]

[0023] A1 is a constant coefficient term of the signal amplitude. The histogram statistics of the two-dimensional image of the amplitude of the compressed signal are made, and the histogram bin size interval is determined by formula (6):

[0024]

[0025] σ is the standard deviation of the entire amplitude image, and n=NKL is the number of pixels in the entire image. The quartile point at the edge of the histogram bin is determined as the threshold. Data points less than the threshold are set to zero, and the remaining data points are kept unchanged. Then, the distance of the target is determined along the peak of each slow time τ according to formula (7):

[0026]

[0027] Finally, an interpolation operation is performed on the original distance information to ensure distance smoothness while removing outliers.

[0028] Step 8: Use the extended Kalman filter algorithm, take the two-dimensional coordinates of the target as the state variable, the two-dimensional speed of the target as the control variable, and the distance measured by the base station as the observation value for iterative processing. Through multiple predictions and updates of state variables and observation values, the system error is gradually corrected, and finally the complete target motion trajectory after filtering and estimation is output.

[0029] After adopting the above-described scheme, the present invention can achieve the following beneficial effects: first, it can be directly deployed on the base station to realize communication perception integration; second, the scheme of mounting the perception receiver on the base station side can reduce the hardware requirements and reduce the hardware system deployment cost; third, the algorithm has low complexity. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 This is a geometric diagram of how the system works.

[0031] Figure 2 Generates a schematic diagram for the particles.

[0032] Figure 3 The data processing flow chart is shown in Figure 2. DETAILED DESCRIPTION

[0033] The first stage is the base station's rough positioning of the drone:

[0034] Step 1: The UAV positioning system is mainly divided into the signal transmission part, the receiving part and the information processing part. The system is deployed on the base station in a side-mounted manner. The radar transceiver antennas in the system are placed in the same position, and the radar works in electronic scanning mode. The transmitting platform turns on the beam scanning working mode, and switches the beam in the pitch and azimuth directions in a fixed period on a pre-set monitoring plane. The receiving platform adopts the fixed beam working mode. The system working concept diagram is as follows Figure 1 shown.

[0035] Step 2: Start the radar transmission module beam scanning, receive the echo signal, and pre-process the echo signal. For a single target, the transmitting antenna transmits the OFDM signal in the millimeter wave frequency band, and the receiving platform receives the echo signal from the target. After removing the carrier frequency, the received echo signal can be expressed as

[0036]

[0037] In formula (8), t is the fast time, τ is the slow time, N is the number of subcarriers, and f c is the carrier frequency, c is the speed of light, λ is the wavelength, Δf is the subcarrier spacing, T is the pulse width, R(τ) is the slant range between the target and the radar, represents the modulation function of the antenna pointing, θ P and Respectively represent the azimuth and elevation angle of the target at that moment, and the rectangular pulse function is

[0038]

[0039] Formula (8) can be expressed as the convolution relationship between the target scattering and the point spread function, which can be expressed in matrix form as

[0040] y=HsZ T (10)

[0041] Where y represents the echo matrix, s represents the target scattering coefficient matrix, H represents the azimuth convolution matrix, Z represents the elevation convolution matrix, and n represents the noise matrix. According to the beam scanning period of K pulses, the echo matrix is ​​divided into a two-dimensional signal sequence y1, y2, ..., y L , L is the number of frames, and the radar pattern matrices H and Z within a scanning cycle are known.

[0042] Solving the target scattering matrix is ​​actually an optimization problem, and the regularized optimization problem is constructed as:

[0043]

[0044] Since two-dimensional linear inversion is difficult to achieve, the matrix vectorization is converted into the inversion form of one-dimensional inversion. The signal is vectorized into s i =vec(s i ), y i =vec(y i ), the convolution matrix represents the Kronecker product of the matrix. For the two-dimensional signal matrix at each time point, the solution of the target optimization function can be obtained by the least squares method:

[0045]

[0046] Where I represents the identity matrix and λ is the regularization parameter, which ensures that the matrix is ​​positive definite so that the inversion result is stable.

[0047] Step 3: Find the peak position of the target scattering coefficient matrix amplitude in equation (12), then the two-dimensional coordinate position of the target is the two-dimensional coordinate position of the target scattering coefficient matrix corresponding to the peak value:

[0048]

[0049] where x i ,y i is the two-dimensional positioning coordinate point of the target in the i-th frame.

[0050] The second stage is the optimization of the drone trajectory based on the base station's rough positioning parameters:

[0051] Step 4: Pre-filter the coarse positioning trajectory, smooth the trajectory and remove unreasonable trajectory points. Apply an average sliding filter to smooth the entire trajectory, reduce the impact of unreasonable trajectory points on the trajectory, and eliminate sharp changes. After eliminating obvious trajectory deviation values, there may still be some unreasonable positioning results in the trajectory. Optimize the trajectory based on the speed threshold method, and set the speed thresholds in the x and y directions respectively. Determine which points are unreasonable trajectory points based on the speed in each direction, and use the average value obtained by the local average method to replace unreasonable trajectory points.

[0052] Step 5: Generate a point cloud around the two-dimensional coordinates of each positioning point obtained by rough positioning. The generated point cloud should be able to cover the error range of radar detection. Randomly select points from each point cloud group to form a trajectory like Figure 2 As shown in the figure, M trajectories are selected as the initial particle swarm, and the particle swarm optimization algorithm is used to optimize the coarse positioning trajectory according to the known target motion model and the root mean square error (RMSE) of polynomial fitting is used as the cost function.

[0053] First, initialize the particles, calculate the loss value of each particle through the loss function, and the particle with the smallest loss value is recorded as the global optimal solution:

[0054]

[0055] Where E(·) represents the fitting error. The P-term fitting method is adopted. The specific value of P depends on the motion model used. Within a certain range, the larger the P, the better the fitting effect of the nonlinear trajectory. Suppose any particle p i The x-coordinate of the trajectory is the vector x i , the y coordinate of the trajectory is the vector y i It can be expressed as

[0056]

[0057] where n i is the fitting error, we can know is the vector x i Vandermonde matrix, the matrix W i Performing QR decomposition can obtain an orthogonal matrix and an upper triangular matrix Make Ignore the fitting error n i When , the coefficient vector is obtained by inverting

[0058]

[0059] The fitting error is

[0060] E(p i )=||n i ||2=||y i -W i b i ||2 (17)

[0061] Then update the speed of each particle. To prevent the particles from migrating to a far distance due to excessive speed, the speed needs to be limited. Then update the position of each particle, and update the local optimal solution and the global optimal solution according to the new position of the particle. Repeat the update of speed and the update of the local optimal solution and the global optimal solution until the global optimal solution is no longer updated after a certain number of iterations, or the iteration count reaches the maximum number of iterations, then stop the algorithm.

[0062] Step 6: Based on the optimized trajectory results, use the position backward difference method to obtain the two-dimensional average motion speed of the moving target The speed of the first position can be obtained through Gaussian process regression: the time series corresponding to the speed and the speed of the two dimensions are used as the training data of the Gaussian probability model. The complete time series is used as the input of the model to obtain the fitted two-dimensional speed sequence. The speed between two positions is considered to be equal to the speed of the previous position.

[0063] Step 7: Extract distance information from the original echo signal. Formula (8) performs FFT along the distance direction to perform distance compression. The compressed signal is in the form of

[0064]

[0065] A1 is a constant coefficient term of the signal amplitude. The histogram statistics of the two-dimensional image of the amplitude of the compressed signal are made, and the histogram bin size interval is determined by formula (19):

[0066]

[0067] σ is the standard deviation of the entire amplitude image, and n = NKL is the number of pixels in the entire image. The quartile point at the edge of the histogram bin is determined as the threshold. Data points less than the threshold are set to zero, and the remaining data points are kept unchanged. Then, the distance of the target is determined along the peak of each slow time τ according to formula (20):

[0068]

[0069] Finally, an interpolation operation is performed on the original distance information to ensure distance smoothness while removing outliers.

[0070] Step 8: Use the extended Kalman filter algorithm to take the two-dimensional coordinates of the target as the state quantity X i , the two-dimensional velocity of the target is used as the control quantity U i , the distance r measured by the base station i As the observed value, its state transfer equation is

[0071]

[0072] Where Δt represents the velocity sampling interval. Its observation equation is:

[0073]

[0074] Repeat the prediction and update steps. Through multiple predictions and updates of state variables and observations, the system error is gradually corrected, and finally the complete target motion trajectory after filtering and estimation is output.

[0075] The entire data processing process is as follows Figure 3 shown.

[0076] The above description is only a specific implementation mode of the present invention. Any feature disclosed in this specification, unless otherwise stated, can be replaced by other alternative features that are equivalent or have similar purposes; all the disclosed features, or all the steps in the methods or processes, except for mutually exclusive features and / or steps, can be combined in any way.

Claims

1. A method for positioning a drone target based on a communication base station beam scanning method, comprising the following steps: Step 1: The system is deployed on the base station in a side-mounted manner. The radar transmitting and receiving antennas in the system are placed in the same position. The radar works in electronic scanning mode. The transmitting platform starts the beam scanning working mode and switches the beams in the elevation and azimuth directions in a fixed period on a pre-set monitoring plane. The receiving platform adopts a fixed beam working mode. The system working concept diagram is shown in Figure 1. Step 2: After the transmitting platform transmits the signal, the receiving platform receives the echo signal reflected by the target and uses the least squares method to solve the target scattering matrix. Step 3: Find the peak position of the target scattering coefficient matrix amplitude in the second step. The two-dimensional coordinate position of the target is the two-dimensional coordinate position of the target scattering coefficient matrix corresponding to the peak value. Step 4: Apply average sliding filtering to smooth the entire trajectory, then optimize the trajectory based on the speed threshold method, and use the average value obtained by the local average method to replace unreasonable trajectory points. Step 5: Generate a point cloud around the two-dimensional coordinates of each positioning point obtained by coarse positioning, and randomly select points from each point cloud group to form a trajectory, as shown in Figure 2. Select multiple trajectories as the initial particle swarm, use the particle swarm optimization algorithm, and use the root mean square error of polynomial fitting as the cost function according to the known target motion model to optimize the coarse positioning trajectory. Step 6: Based on the trajectory results obtained by optimization, use the backward difference method to obtain the two-dimensional average speed of the moving target. Use the time series corresponding to the average speed and the average speed as the training data for Gaussian process regression. Input the complete time series prediction to get the speed of the first position. The speed between the two positions is considered to be equal to the speed of the previous position. Step 7: Perform FFT along the distance direction to perform distance compression, make histogram statistics on the two-dimensional image of the amplitude of the compressed signal, determine the quartile point on the edge of the histogram bin as the threshold, set the data points less than the threshold to zero, keep the remaining data points unchanged, find the peak along each slow time dimension of the compressed signal to determine the target distance, and finally perform an interpolation operation on the original distance information obtained. Step 8: Use the extended Kalman filter algorithm, take the two-dimensional coordinates of the target as the state variable, the two-dimensional speed of the target as the control variable, and the distance measured by the base station as the observation value, and perform iterative processing. Through multiple predictions and updates of the state variables and observation values, the system error is gradually corrected, and finally the complete target motion trajectory after filtering and estimation is output.

Images