A Fast Distributed MIMO Radar Multi-Target Detection and Localization Method

Abstract

The present invention discloses a fast distributed MIMO radar multi-target detection and positioning method, which is applied to the technical fields of distributed radar target detection and positioning. Aiming at the problem of low target detection efficiency existing in the prior art when facing a large number of targets, the present invention first approximates the multi-target high-dimensional joint estimation equation as a low-dimensional estimation equation; then proposes a detector based on the number of channels to achieve low-dimensional multi-target detection; then uses a density-based clustering algorithm to determine the number of targets in the scene and distinguish different targets; finally, finds the position where the maximum value of different clustering results is located to achieve multi-target positioning. The present invention can achieve fast multi-target detection and positioning while ensuring high detection probability and positioning accuracy.

Description

Technical Field

[0001] The present invention belongs to the technical field of distributed radar target detection and positioning, and particularly relates to a multi-target joint direct positioning technology. Background Art

[0002] Distributed multiple-input multiple-output (MIMO) radar receives electromagnetic waves reflected by a target from different directions, and uses spatial diversity gain to effectively reduce the influence of the scintillation characteristics of the target radar cross section (RCS). Therefore, distributed MIMO radar is widely used in industrial perception control, urban digital security, and intelligent vehicle integrated electronic systems. Currently, there are two common positioning methods. The first positioning method is the two-step positioning method, and the method is as follows: each receiver first obtains target measurements, and then the fusion center synthesizes the measurement results of all receivers to achieve target positioning. This method has a fast operation speed, but the detection robustness and positioning accuracy of the target are relatively low. The second positioning method is the direct positioning method, and the method is as follows: the receiver directly transmits the original data to the fusion center, and the fusion center constructs a cost function to obtain the target position. This method has high detection robustness and positioning accuracy, but the solution of the cost function requires a large amount of computational loss, which is not conducive to the requirements of real-time positioning. Therefore, it is of great application value to find a fast multi-target direct positioning algorithm.

[0003] Many research institutions at home and abroad have carried out research on direct positioning methods. Tel Aviv University in Israel proposed a high-dimensional direct positioning method for distributed MIMO radar (Bar-Shalom O, Weiss AJ. Direct positioning of stationary targets using MIMO radar[J]. Signal Processing, 2011, 91(10): 2345-2358.), which deduced the cost functions in the cases of known signals and unknown signals according to the maximum likelihood estimation criterion. However, this method requires prior knowledge of the number of targets in the scene and is difficult to apply in practice. The University of Electronic Science and Technology of China proposed a distributed MIMO radar positioning method based on the CLEAN idea (Yi W, Zhou T, Ai Y, etc. Suboptimal Low Complexity Joint Multi-Target Detection and Localization for Non-Coherent MIMO Radar With Widely Separated Antennas[J]. IEEE Transactions on Signal Processing, 2020, 68: 901-916.), which processes the targets in the scene one by one. After detecting / localizing a target, the influence of this target is eliminated, and the process is repeated until there are no targets in the scene. However, when the number of targets is large, this method still requires a large amount of computing space. From the above methods, it can be seen that the applicable scope of traditional direct positioning methods is limited, and the equipment is required to have high computing power. Therefore, it is of great value to study a fast multi-target direct positioning method for distributed MIMO radar. Summary of the Invention

[0004] To solve the above technical problems, the present invention proposes a fast multi-target detection and positioning method for distributed MIMO radar, which can effectively reduce the problem of excessively high computational complexity brought by maximum likelihood estimation and achieve multi-target detection and positioning with low computational loss.

[0005] The technical solution adopted by the present invention is: a fast multi-target detection and positioning method for distributed MIMO radar, and the application scenario is specifically: in the two-dimensional area detectable by the radar Among them, there are K transmitters, L receivers, and P targets in total. The receivers receive the signals scattered by all the transmitters via all the targets, and there are L received signals in total. Assuming that the signals transmitted by each transmitter are orthogonal to each other, therefore, each received signal can be separated into K reconstructed signals after passing through the matched filters composed of different transmitted signals. Finally, M = K·L reconstructed signals are obtained. After signal sampling, the m-th reconstructed signal is denoted as r m ;

[0006] The multi-target detection and positioning method includes the following steps:

[0007] S1. The high-dimensional multi-target joint position estimator obtained by using maximum likelihood estimation is:

[0008]

[0009]

[0010] Among them, Θ is the concatenation of the target position vectors; λ is the detection threshold; is the estimated value of the complex reflection coefficient vector of the m-th reconstructed signal, is the log-likelihood ratio after substituting the estimated value of the complex reflection coefficient;

[0011] S2. Dimension reduction is performed on the high-dimensional multi-target joint position estimator in step S1 to obtain the dimension-reduced maximum likelihood estimator of the multi-target position as:

[0012]

[0013] Among them, represents adjusting the independent variable A to make f(A) obtain the maximum value;

[0014] S3. The two-dimensional region is divided into X and Y grid points along two perpendicular directions x and y respectively, then there are X×Y grid points in the region; calculate the estimated value of each grid point;

[0015] S4. Use the channel number difference to detect the real targets among all the maximum values and eliminate the "ghost targets";

[0016] S5. Extract the positions of the real targets obtained in step S4 into the target position set;

[0017] S6. Cluster the target position set, and after clustering, find the position of the maximum value of each category, which is the positioning position of the target.

[0018] Advantages of the present invention: The method of the present invention can effectively reduce the problem of excessively high computational complexity caused by maximum likelihood estimation, and achieve multi-target detection and positioning with low computational loss. Compared with the method mentioned in the technical background, this method does not require prior knowledge of target information in the scene and has a lower computational complexity than traditional methods. It ensures the real-time detection and positioning requirements of distributed MIMO radar, and provides a strong guarantee for operators to make correct decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a flowchart for multi-target detection / localization processing.

[0020] Figure 2 It is a schematic diagram of a distributed MIMO radar scenario.

[0021] Figure 3 It is a schematic diagram of ghost targets and real targets.

[0022] Figure 4 It is a schematic diagram of a simulated two-dimensional scenario.

[0023] Figure 5 It is a simulation diagram of grid traversal using the likelihood function.

[0024] Figure 6 It is the simulation result after passing through the detector based on the number of channels.

[0025] Figure 7 It is the result after clustering using the DBSCAN algorithm.

[0026] Figure 8 It is the simulation diagram of the final positioning result.

[0027] Figure 9 It is a curve graph of the detection probability of three targets varying with the signal-to-noise ratio.

[0028] Figure 10 It is a curve graph of the root mean square error of three targets varying with the signal-to-noise ratio.

[0029] Figure 11 It is a comparative curve graph of the detection probabilities of three different methods.

[0030] Figure 12 It is a comparative graph of the normalization rates of three methods. DETAILED DESCRIPTION OF THE INVENTION

[0031] To facilitate those skilled in the art to understand the technical content of the present invention, the following further explains the content of the present invention with reference to the accompanying drawings.

[0032] The present invention proposes a fast distributed MIMO radar detection and positioning method. First, the multi-target high-dimensional joint estimation equation is approximated as a low-dimensional estimation equation; then, a detector based on the number of channels is proposed to achieve low-dimensional multi-target detection; after that, a density-based clustering algorithm is used to determine the number of targets in the scene and distinguish different targets; finally, the maximum values of different clustering results are found to achieve multi-target positioning.

[0033] The technical solution of the present invention is as follows:

[0034] The present invention proposes a fast distributed MIMO radar detection and positioning method, and the processing flow is as Figure 1 shown, including the following steps:

[0035] Step 1: Dimension reduction

[0036] Assume that in the two-dimensional area detectable by the radar, there are a total of K transmitters, L receivers and P targets (as Figure 2 shown). The receivers receive the signals scattered by all the transmitters via all the targets, so there are a total of L received signals. Assume that the signals transmitted by each transmitter are orthogonal to each other. Therefore, after each received signal passes through the matched filters composed of different transmitted signals, it can be separated into K reconstructed signals. Finally, M = K·L reconstructed signals can be obtained. After signal sampling, the m-th reconstructed signal is denoted as r m . The high-dimensional multi-target joint position estimator obtained using Maximum Likelihood Estimation (MLE) is:

[0037]

[0038]

[0039] where, Θ = [[x1,y1] T ,[x2,y2] T ,...,[x P ,y P T T is the concatenation of the target position vectors; λ is the detection threshold; is the estimated value of the complex reflection coefficient vector of the m-th reconstructed signal, is the log-likelihood ratio after substituting the estimated value of the complex reflection coefficient, and its expression is:

[0040]

[0041] where, R m represents the autocorrelation matrix of the noise; (·) H represents the conjugate transpose operation; (·)​​-1 denotes the inverse operation; S m =[s m1 , s m2 ,..., s mP is an N T ×P matrix, where N T represents the length of the discrete sampling signal vector, and s mp represents the size of the m-th reconstructed signal scattered by the p-th target, which is an N T ×1 vector. The n-th element of s mp is:

[0042] s mp [n]=s kp (nT s - τ mp (θ p )) for n = 0, 1,..., N T - 1

[0043] where T s is the sampling time interval; θ p =[x p , y p T is the position of the p-th target; s k (·) is the time-domain waveform of the k-th transmitted signal, and τ mp (θ p ) represents the time delay of the transmitted signal reflected by the p-th target to the receiver.

[0044] Assuming that the reconstructed signal satisfies the narrowband pulse form, the estimated value of the complex reflection coefficient can be reduced in dimension to:

[0045]

[0046] Substituting the estimated value of the complex reflection coefficient back in, the log-likelihood ratio is:

[0047]

[0048] The maximum likelihood estimator of the multi-target position after the final dimensionality reduction is:

[0049]

[0050] where denotes adjusting the independent variable A to make f(A) reach the maximum value; Since it is difficult to obtain an analytical solution for this equation, the two-dimensional region is divided into X and Y grid points along two perpendicular directions x and y respectively. Then there are X×Y grid points in the region. By calculating the estimated values of each grid point, all the maximum values in the two-dimensional region can be found.

[0051] Step 2: Set up a detector based on the number of channels

[0052] However, as Figure 3 shown, since the channels are non-coherently superposed, the maximum value appears not only at the target position but also at the non-coherent superposition of channels. These targets will cause false alarms in detection and are "ghost targets". However, since the echo signal is focused at the real target, the number of high-energy channels at the real target position is much higher than that at the "ghost target" position. Therefore, the difference in the number of channels can be used to detect real targets and eliminate "ghost targets". Detect each channel, and the structure of the channel detector at position θ is as follows:

[0053]

[0054] where λ m,local is the detection threshold of the m-th reconstructed signal, which determines the detection probability and false alarm probability of the detector.

[0055] After detection, add up the results of all detectors to obtain the number of channels detected by the channels:

[0056]

[0057] The purpose of the global detector is to determine whether the target at position θ is a real target. If the number of channels exceeds the global detector threshold, it is considered a real target and the estimated value is retained; if the number of channels is lower than the global detector threshold, it is considered that there is no target and the estimated value is deleted. Therefore, the expression of the global detector is:

[0058]

[0059] where λ global represents the global detection threshold, which depends on the required performance level. Considering that the RCS of the target has a scintillation characteristic and the "ghost target" is formed by at least 2-channel non-coherent superposition, the threshold is set here to half detection, that is

[0060] Step 3: DBSCAN clustering and search for the maximum value of the clustering result

[0061] Extract the position θ passing through the detector into the target position set Ψ, and we have:

[0062]

[0063] After passing through the detector based on the number of channels, the true target positions will retain the estimated values, while the other positions will be set to zero. Therefore, the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) can be used to classify different targets. The calculation formula of DBSCAN is:

[0064] I = DBScan(Ψ, ε, Pts)

[0065] where DBScan(·) represents the execution of the DBSCAN algorithm; ε is the scanning radius; Pts is the minimum number of points within the scanning radius; I is the class index output by the algorithm, and the elements in it correspond one-to-one with the elements in Ψ.

[0066] After clustering, assuming there are P' classes in I, that is, I = {1,..., 1, 2,..., 2,..., P',..., P'}. Then the target position set Ψ can be divided into P' subsets, and the p'-th subset is:

[0067] ψ p' = Ψ(I = p'), p' = 1, 2,..., P'

[0068] where Ψ(I = p') represents the set of elements with index p' in the set Ψ.

[0069] According to the properties of the set, it is easy to obtain:

[0070] ψ1 + ψ2 +... + ψ P' = Ψ

[0071] Each class can be regarded as an extended target. Therefore, finding the maximum estimated value of each class is the positioning position of the target. The positioning result of the p'-th class is:

[0072]

[0073] Finally, the positioning results of all targets are:

[0074]

[0075] The simulation scenario is as Figure 4As shown in the figure, there are 3 targets in the two-dimensional scene, and the coordinates of the three targets are located at: (4.8, 3.4), (5.5, 5.7), (4.2, 8.1). The radar array is arranged on the x-axis, with 4 transmitters and 5 receivers. The coordinates of the transmit array are located at: (2, 0), (4, 0), (6, 0), (8, 0); the coordinates of the receive array are located at: (0, 0), (2.5, 0), (5, 0), (7.5, 0), (10, 0). The transmitted signal of the radar is a Gaussian pulse signal with a center frequency of 0.8 GHz and a 3 dB bandwidth of approximately 654 MHz. The three targets satisfy a zero-mean complex Gaussian distribution with the same energy intensity. The noise in the two-dimensional space is uniform noise and satisfies a zero-mean complex Gaussian distribution. Therefore, the local threshold of each channel is the same, i.e., λ 1,local = λ 2,local = … = λ M,local . The false alarm probability is set to 1%, and 1000 Monte Carlo simulations are performed at each signal-to-noise ratio. The scanning radius of the DBSCAN algorithm is set to the range resolution of the radar, and the number of points within the scanned area is set to 5.

[0076] Processing steps according to the present invention:

[0077] Step 1: Dimension reduction

[0078] The two-dimensional scene is divided into grids in the x-direction and y-direction. The grid resolution in both the x-direction and y-direction is 0.1 m. Then, the (10×10) m 2 two-dimensional scene of size is divided into a total of 101×101 grid points. All grid points are calculated according to the maximum likelihood estimation formula after dimension reduction, and Figure 5 is obtained.

[0079] Step 2: Set a detector based on the number of channels

[0080] The local threshold is adjusted by the false alarm probability and has different values at different signal-to-noise ratios. The specific values are shown in Table 1.

[0081] Table 1 Thresholds corresponding to different signal-to-noise ratios

[0082] Signal-to-noise ratio 0 1 2 3 4 5 6 7 Threshold 2.88 2.79 2.61 2.50 2.40 2.30 2.19 2.10 Signal-to-noise ratio 8 9 10 11 12 13 14 15 Threshold 1.985 1.895 1.82 1.75 1.69 1.63 1.55 1.50

[0083] Since the radar is a 4-transmitter and 5-receiver radar, the global threshold is 10. Let the estimation results at all positions in the two-dimensional space pass through the detector, and Figure 6 can be obtained.

[0084] Step 3: DBSCAN clustering and search for the maximum value of the clustering result

[0085] After clustering using the clustering algorithm, the result can be obtained as shown in Figure 7From the clustering results shown, it can be seen that different energy clusters distributed in the two-dimensional scenario are divided into different categories. According to the positioning formula, the positions where the maximum values of each category are located are found respectively, and finally the positioning result of the target is obtained as Figure 8 shown.

[0086] To verify the feasibility of this method, the present invention also conducts 1000 Monte Carlo simulation experiments. After the Monte Carlo simulation, the detection probability curve and the positioning error curve are as Figure 9 、 10 shown. It can be seen from the detection probability curve that the method proposed by the present invention has good detection performance under high signal-to-noise ratio (SNR, SIGNAL-NOISE RATIO); it can be seen from the positioning error (i.e., the RMSE in Figure 10 , Root Mean Square Error) curve that the method proposed by the present invention can maintain good positioning accuracy under all signal-to-noise ratios, and the worst positioning error does not exceed 0.12m.

[0087] To illustrate the advantages of the method of the present invention, the present invention is also compared with two traditional methods, and the detection curve and the normalized speed are as Figure 11 、 12 shown. For the convenience of description, the two traditional methods are named: CLEAN method (Yi W, Zhou T, Ai Y, etc. Suboptimal Low Complexity Joint Multi-Target Detection andLocalization for Non-Coherent MIMO Radar With Widely Separated Antennas[J]. IEEE Transactions on Signal Processing, 2020, 68: 901-916.), energy detector method (AiY, Yi W, Cui G, et al. Multi-target localization for noncoherent MIMO radar withwidely separated antennas[C]. 2014IEEE Radar Conference. 2014: 1267-1272.). It can be seen from Figure 11 that the detection probability of the method of the present invention is worse than that of the CLEAN method, but better than that of the energy detector method. It can be seen from Figure 12 that the operation speed of the method of the present invention is basically the same as that of the energy detector method, but exceeds the speed of the CLEAN method by about 4 times.

[0088] As can be seen from the simulation results, the method provided by the present invention can achieve multi-target rapid detection and positioning.

[0089] Those of ordinary skill in the art will realize that the embodiments described herein are to assist the reader in understanding the principles of the present invention, and it should be understood that the scope of protection of the present invention is not limited to such specific statements and embodiments. For those skilled in the art, various modifications and variations can be made to the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the scope of the claims of the present invention.

Claims

1. A fast distributed MIMO radar multi-target detection and positioning method, characterized in that The application scenario is specifically as follows: in the two-dimensional area detectable by the radar there are a total of K transmitters, L receivers and P targets. The receivers receive the signals scattered by all the transmitters via all the targets, and there are L received signals in total. Assuming that the signals transmitted by each transmitter are orthogonal to each other, therefore, after each received signal passes through the matched filters composed of different transmitted signals, it can be separated into K reconstructed signals; finally, M = K·L reconstructed signals are obtained; after signal sampling, the m-th reconstructed signal is denoted as r m ; The multi-object detection and positioning method includes the following steps: S1. The high-dimensional multi-object joint position estimator obtained by using maximum likelihood estimation is: where Θ is the concatenation of the target location vectors; λ is the detection threshold; is the estimated value of the complex reflection coefficient vector of the m-th reconstructed signal, is the log-likelihood ratio after substituting the estimated value of the complex reflection coefficient; S2. The high-dimensional multi-object joint position estimator in step S1 is dimensionally reduced to obtain the dimensionally reduced maximum likelihood estimator of the multi-object position as: Among them, denotes adjusting the independent variable A to make f(A) achieve the maximum value; θ denotes the target position; S3. Divide the two-dimensional region into X and Y grid points in two perpendicular directions x and y respectively, so there are X×Y grid points in the region; calculate the estimated value of each grid point; S4. Detect real targets among all maxima by using the channel number difference and eliminate "ghost targets"; S5. Extract the positions of the real targets obtained in step S4 into the target position set; S6. Cluster the target position set, and after clustering, find the position of the maximum value of each category, which is the positioning position of the target.

2. A fast distributed MIMO radar multi-target detection and positioning method according to claim 1, characterized in that The dimensional reduction described in step S2 is specifically as follows: S21. Assume that the reconstructed signal satisfies the narrowband pulse form, and the estimated value of the complex reflection coefficient is dimensionally reduced to: S22. According to the estimated value of the complex reflection coefficient dimensionally reduced in step S21, the updated log-likelihood ratio is obtained as: S23. According to the updated log-likelihood ratio in step S22, the dimensionally reduced maximum likelihood estimator of the multi-object position is obtained as:

3. A fast distributed MIMO radar multi-target detection and positioning method according to claim 2, characterized in that, Step S4 is specifically as follows: S41. Detect each channel. The structure of the channel detector at position θ is: where λ m,local is the detection threshold of the m-th reconstructed signal; S42. Add the results of all detectors to obtain the number of channels detected by the channels: S43. Use a global detector to judge whether the target at position θ is a real target. The expression of the global detector is: Among them, λ global represents the global detection threshold.

4. A fast distributed MIMO radar multi-target detection and positioning method according to claim 3, characterized in that In step S43 represents rounding up.

5. A fast distributed MIMO radar multi-target detection and positioning method according to claim 4, characterized in that, Step S6 specifically uses a density-based clustering algorithm, and the calculation formula is: I = DBScan(Ψ, ε, Pts) where DBScan(·) represents executing the DBSCAN algorithm; Ψ represents the target position set; ε is the scanning radius; Pts is the minimum number of points within the scanning radius; I is the category index output by the algorithm, and the elements therein correspond one by one to the elements in Ψ.

Images