Target recognition and tracking method based on cognitive FDA-MIMO radar network

Abstract

The present invention discloses a target recognition and tracking method based on a cognitive FDA-MIMO radar network, comprising the following steps: S1: establishing a measurement model of radar network targets and mainlobe deceptive trajectory interference; S2: using mainlobe deceptive trajectory interference and the distribution characteristics of targets in space relative to different radars to perform FDA-MIMO radar network target recognition; S3: extracting target measurement information from the target recognition results of each radar and initializing the target state; S4: predicting the target state for each radar and updating the corresponding target state; S5: fusing the target states obtained by each radar to obtain the final target state; S6: repeating steps S4 and S5 to achieve closed-loop cognitive target tracking. This method can accurately identify real targets and mainlobe deceptive trajectory interference in a mainlobe deceptive trajectory interference environment, and initialize a tracker when the target prior information is unknown, thereby achieving target tracking.

Description

Technical Field

[0001] The present invention belongs to the field of signal processing technology, and in particular relates to a target recognition and tracking method based on a cognitive FDA-MIMO radar network. Background Art

[0002] Frequency Diverse Radar (FDA) radar is a newly emerging radar system in recent years. Unlike phased array (PA) and multiple-input multiple-output (MIMO) radar systems, FDA radar systems introduce a slight frequency difference between each transmitting element, making the FDA radar's transmission pattern dependent not only on angle but also on range. Combining FDA and MIMO radars, FDA-MIMO radars not only benefit from the spatial diversity of MIMO radars but also offer controllable degrees of freedom in the range dimension. Early research on FDA-MIMO radars focused primarily on static targets, such as parameter estimation, clutter suppression, and interference (especially mainlobe interference) identification and suppression. In particular, interference suppression struggles due to the lack of range-dependent controllable degrees of freedom in traditional PA and MIMO radars, making them difficult to combat. FDA-MIMO radars, on the other hand, leverage this controllable degree of freedom to suppress mainlobe interference. However, it's worth noting that existing methods for suppressing mainlobe range deception jamming using FDA-MIMO radars rely on two conditions: first, the jamming signal must be delayed by at least one pulse period relative to the target signal, and second, prior target range information must be available or the number of ambiguities known. In reality, these two assumptions are difficult to achieve in actual combat environments. The difficulty in obtaining prior target information remains a pressing issue.

[0003] Research on FDA-MIMO radars for dynamic target tracking is relatively limited. Researchers first applied cognitive FDA-MIMO radars to range-angle transmit beamforming with low probability of intercept (LPI). Subsequently, the subarray concept was extended to cognitive FDA-MIMO radars, and cognitive target tracking was achieved by optimizing the transmit spatial matrix using the maximum SNR and minimization of the Clamer-Rao lower bound. However, this approach only considers interference-free scenarios. For scenarios with interference, existing techniques have proposed a joint transmit / receive beamforming design method for cognitive FDA-MIMO radars based on the maximum SNR principle for interference suppression and target tracking. However, prior information about the target is still required to distinguish between the target and the interference, especially during the tracker initialization phase. Furthermore, another existing approach proposed a cognitive FDA-MIMO radar system with energy-efficient adaptive transmit beamspace design in interference environments. However, this interference and target identification method only addresses static or slow-moving interference and does not consider situations where the interference has a regular trajectory.

[0004] The rapid development of electronic jamming technology, particularly the emergence and development of digital radio frequency memory (DFM), has greatly facilitated the application of deceptive jamming on the battlefield, posing a more severe challenge to the battlefield survivability of modern radars. Jammers utilize DRFM (digital radio frequency memory) to intercept, store, modulate, and forward enemy radar signals, creating highly realistic signals that mimic the real target in the time-space-frequency domain, thereby deceiving enemy radars. In battlefield environments, jammers can instantaneously modulate and capture radar signals and forward them. However, the delay between the jamming signal and the target signal cannot be guaranteed to exceed one pulse. Because there is no delay of more than one pulse between the jamming and target signals, the target ambiguity cannot be determined in advance. Using the fast-time delay to determine the target's prior range information is unreliable because the target ambiguity is unknown. Furthermore, the deceptive targets forwarded by the jammer can form a series of realistic false target tracks in the slow-time domain (with a regular speed). Consequently, traditional FDA-MIMO radar methods for identifying and suppressing mainlobe deceptive jamming are completely ineffective. Consequently, target detection and tracking using a single radar is no longer viable in today's complex jamming environment.

[0005] In summary, in the face of mainlobe deceptive trajectory jamming environment, traditional FDA-MIMO radar recognition has difficulty distinguishing mainlobe deceptive trajectory jamming from real targets; and because the target prior information is unknown, it is also difficult to initialize the tracker, making it impossible to achieve target tracking. Summary of the Invention

[0006] To address the above-mentioned problems in the prior art, the present invention provides a target recognition and tracking method based on a cognitive FDA-MIMO radar network. The technical problem to be solved by the present invention is achieved through the following technical solutions:

[0007] A target recognition and tracking method based on a cognitive FDA-MIMO radar network, comprising:

[0008] S1: Establish a measurement model for radar net targets and mainlobe deceptive trajectory jamming;

[0009] S2: Based on the measurement model, the FDA-MIMO radar network is used to identify targets using mainlobe deceptive trajectory jamming and the spatial distribution characteristics of targets relative to different radars to obtain target recognition results for each radar;

[0010] S3: Extract the target measurement information from the target recognition results of each radar, and initialize the target state based on the G-pair large-interval measurement drive;

[0011] S4: Use the EKF tracker and the initialized target state to predict the target state of each radar and update the corresponding target state;

[0012] S5: Use the track fusion algorithm to fuse the target states obtained by each radar to obtain the final target state;

[0013] S6: Repeat the operations of steps S4 and S5 to achieve closed-loop cognitive target tracking and output the target tracking trajectory.

[0014] Beneficial effects of the present invention:

[0015] 1. The target recognition and tracking method based on cognitive FDA-MIMO radar network provided by the present invention first establishes the signal model of mainlobe deceptive trajectory interference and target, then identifies the target and mainlobe deceptive trajectory interference by utilizing FDA-MIMO radar networking, and finally uses the initialized target state for cognitive tracking. This method can accurately identify the real target and the mainlobe deceptive trajectory interference in the mainlobe deceptive trajectory interference environment, and the time delay between the interference and the target can be less than one pulse. In addition, when the target prior information is unknown, the tracker is initialized to achieve target tracking.

[0016] 2. The present invention adopts the target state initialization method of G-pair large interval measurement drive to better ensure stable tracking;

[0017] 3. The present invention uses the optimal weight solution algorithm of the maximum Capon power spectrum criterion to suppress the main lobe deceptive trajectory interference and extract the real target measurement information, which can ensure that the main lobe of the pattern always points to the distance and angle where the target is located, avoid the divergence of subsequent tracking, and improve the tracking accuracy;

[0018] 4. After obtaining the updated target state, the present invention also uses the target state correction algorithm based on auxiliary particles to perform correction, effectively avoiding the tracking divergence problem caused by the inaccurate initialization parameters of the traditional EKF tracker;

[0019] 5. The present invention improves the performance of a single FDA-MIMO radar with large target tracking error after networking through a track fusion strategy, thereby improving the overall tracking effect.

[0020] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 A system architecture diagram for target recognition and tracking of a cognitive FDA-MIMO radar network provided by an embodiment of the present invention;

[0022] Figure 2 A schematic diagram of main lobe trajectory interference provided by an embodiment of the present invention;

[0023] Figure 3 A schematic diagram of mainlobe deceptive trajectory interference and target distribution in an FDA-MIMO radar network provided by an embodiment of the present invention;

[0024] Figure 4 A schematic diagram of a target and main lobe deceptive trajectory jamming provided by an embodiment of the present invention;

[0025] Figure 5 A flowchart of a target recognition and tracking method based on a cognitive FDA-MIMO radar network provided in an embodiment of the present invention;

[0026] Figure 6 A schematic diagram of target tracking results of a cognitive FDA-MIMO radar network provided by an embodiment of the present invention;

[0027] Figure 7 A comparison chart of the RMSE error of target position estimation for radar 1 in the cognitive FDA-MIMO radar network target tracking using different algorithms provided in an embodiment of the present invention;

[0028] Figure 8 A comparison chart of the RMSE error of target position estimation for Radar 2 in the cognitive FDA-MIMO radar network target tracking using different algorithms provided in an embodiment of the present invention;

[0029] Figure 9 A comparison chart of the RMSE error of target velocity estimation for Radar 1 in the cognitive FDA-MIMO radar network target tracking using different algorithms provided in an embodiment of the present invention;

[0030] Figure 10 A comparison chart of the RMSE error of target velocity estimation of Radar 2 in the cognitive FDA-MIMO radar network target tracking under different algorithms provided by an embodiment of the present invention;

[0031] Figure 11 A comparison diagram of the output signal-to-interference-and-noise ratio of radar 1 during target tracking in a cognitive FDA-MIMO radar network under different algorithms provided by an embodiment of the present invention;

[0032] Figure 12 A comparison diagram of the output signal-to-interference-and-noise ratio of radar 2 during target tracking in a cognitive FDA-MIMO radar network under different algorithms provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0033] The present invention will be further described in detail below with reference to specific examples, but the embodiments of the present invention are not limited thereto.

[0034] Example 1

[0035] The present invention constructs an FDA-MIMO radar network system in the main lobe deception trajectory jamming environment and adopts a cognitive approach to target recognition and tracking. Figure 1 , Figure 1The system architecture diagram of target recognition and tracking of a cognitive FDA-MIMO radar network provided by an embodiment of the present invention. Its workflow is divided into two stages. (1) Search stage: In the search stage, the distance and angle information of the target are unknown, and there is main lobe trajectory interference. At this time, first, multiple FDA-MIMO radars located in different positions are used to search the airspace separately, and the distance and angle estimation of the target and interference is realized by using the distance-angle joint estimation method; secondly, based on the spatial distribution characteristics of the real target and the main lobe trajectory interference of different radars, the distance and angle information of the target estimated by different radars are homologously detected to distinguish the real target from the interference; finally, the track of the real target is initialized using the identified target and interference information. (2) Tracking stage: Once the target track initialization is completed, all radars enter the tracking mode. In tracking mode, the target's track is first predicted using an EKF predictor. Each radar then transmits a signal based on the predicted target position. At the receiver, a minimum variance distortionless detector (MVDR) is used to determine optimal weights. The optimal weights are then used to identify the target on a single FDA-MIMO radar, and the resulting actual target measurement information is used to update the target state predicted by each radar. Finally, the updated target track information from each radar is fused at the track fusion center to reduce individual radar tracking errors. After trajectory fusion is complete at the track fusion center, the fused target position information is retransmitted to the EKF (Extended Kalman Filter) predictor, entering the next CPI (Coherent Processing Interval) cycle.

[0036] For details, see Figure 2 , Figure 2 Schematic diagram of a mainlobe deceptive trajectory jammer and a target. The jammer is located in the mainlobe direction, differing only in distance from the target. Through time delay and Doppler modulation, the deceptive target can lag behind or lead the real target. Furthermore, the deceptive target echo signal and the real target echo signal may be within the same pulse or separated by more than one pulse. Traditional mainlobe anti-jamming algorithms can only counter deceptive targets with a delay of more than one pulse.

[0037] Furthermore, taking two radars as an example, Figure 3 and Figure 4 , Figure 3 This figure illustrates the distribution of mainlobe deceptive trajectory jamming and targets in an FDA-MIMO radar network, according to an embodiment of the present invention. Clearly, the mainlobe range deceptive jamming of Radar 1 and Radar 2 is spatially discrete, while the true targets overlap. This means that true targets can be distinguished from false targets using information-level homology detection. Figure 4The trajectories of the real target and the main lobe deceptive trajectory jammer set in the numerical experiment are given. The false target is always in the same direction as the real target and has a regular speed to deceive the radar.

[0038] In this embodiment, a target identification and tracking method based on a cognitive FDA-MIMO radar network is proposed within the aforementioned FDA-MIMO radar network system. This method is used for target identification and tracking in a mainlobe deceptive trajectory interference environment. The specific concepts are as follows: First, in the presence of mainlobe deceptive trajectory interference, signal models for targets and mainlobe deceptive trajectory interference within the FDA-MIMO radar network are developed. Second, the distribution characteristics of the target and mainlobe deceptive trajectory interference on different radars are used to identify the target and interference. A target state initialization method driven by G-pair large-interval measurements is proposed to initialize the target state in the EKF tracker. Third, during the cognitive tracking process, a most weighted solution method based on the maximum Capon power spectrum criterion is proposed for interference suppression and target measurement information acquisition for a single radar. Fourth, an auxiliary particle-based target state correction algorithm is proposed to overcome the tracking divergence problem caused by inaccurate initialization parameters in traditional EKF filters. Finally, a cognitive approach is adopted to continuously provide the radar with the latest environmental information using the tracker, achieving cognitive tracking of the target.

[0039] The target recognition and tracking method based on cognitive FDA-MIMO radar network provided by this embodiment is described in detail below. Figure 5 , Figure 5 A flowchart of a target recognition and tracking method based on a cognitive FDA-MIMO radar network provided in an embodiment of the present invention includes:

[0040] S1: Establish a measurement model for radar net target and mainlobe deceptive trajectory jamming.

[0041] In this embodiment, assuming that the established FDA-MIMO radar network has H radars, the measurement model of the target and mainlobe deceptive trajectory jamming of the hth (h=1, 2, ..., H)th radar in the radar network is:

[0042]

[0043]

[0044] Where Y h,k represents the target measurement model of the hth radar in the kth CPI, Represents the target measurement of the hth radar within the kth CPI of the ith range gate, 1≤i≤N bin , N bin Indicates the number of range gates; represents the main lobe deceptive trajectory jamming measurement model of the kth CPI of the hth radar, and They are the nth generation of target and main lobe deceptive trajectory interference at the kth CPI after down-conversion matched filtering at the receiving end. bin The signal on the distance gate, All elements in the are subject to mean 0 and covariance σ h 2 The complex white Gaussian noise component of .

[0045] Ignoring the CPI index k, the signal model of the target and main lobe deceptive trajectory interference is:

[0046]

[0047]

[0048] In the formula, the superscript (·) T represents transpose, is the Kronecker product, a(·,·), b(·) and a d (·) denote the transmit steering vector, receive steering vector and Doppler steering vector, ξ and ξ respectively. J denote the backscatter coefficient of the target and the backscatter coefficient of the Jth main lobe deceptive trajectory jammer, r h ,θ h and f h,d are the distance, angle and Doppler shift of the target relative to the hth radar, r h,J ,θ h,J and f h,d,J are the distance, angle and Doppler frequency shift of the Jth main lobe deceptive track jammer relative to the hth radar, respectively. (J) Indicates a total of N (J) A disturbance.

[0049] Ignoring the interference index J, the transmission steering vector a(·,·), the receiving steering vector b(·) and the Doppler steering vector a of the hth radar are d (·) can be expressed as:

[0050]

[0051]

[0052]

[0053] Among them, π represents the unit of pi, j is the imaginary unit, c represents the speed of light, M h and N hdenote the number of transmitting and receiving array elements of the hth FDA-MIMO radar, Δf h is the frequency difference of the signals transmitted by adjacent transmitting elements of the hth radar, f h,0 is the reference frequency of the transmitted signal. The spacing between adjacent array elements in the transmitting and receiving ends is half a wavelength d h , T CPI It represents the length of a coherent integration time, and L represents the number of pulses accumulated within a coherent integration time.

[0054] S2: Based on the measurement model, the FDA-MIMO radar network is used to identify targets using mainlobe deceptive trajectory interference and the distribution characteristics of targets in space relative to different radars to obtain target recognition results for each radar.

[0055] In this embodiment, step S2 includes:

[0056] S21: Calculate the number of each radar at the nth bin Capon power spectrum within the k-1th CPI of a range gate.

[0057] Specifically, suppose the hth radar receives the nth bin Signal on the range gate Then the hth radar is at the nth bin The signal + interference + noise covariance matrix within the k-1th CPI of the range gate is:

[0058]

[0059] Superscript (·) H Represents the conjugate transpose operator.

[0060] Then the hth radar is at the nth bin The Capon power spectrum within the k-1th CPI of a range gate can be formulated as:

[0061]

[0062] in, is the joint transmit-receive virtual steering vector, R h,u Indicates the maximum unambiguous range of the hth radar.

[0063] Therefore, the Capon power spectrum on all range gates can be expressed as:

[0064]

[0065] Among them U R {·} represents the union along the range gate direction.

[0066] S22: Using the obtained Capon power spectrum and the CFAR detector, the distance and angle information of the target and interference of each radar at the k-1th CPI is extracted to obtain the measurement set of all radars.

[0067] Specifically, under high signal-to-noise ratio and high interference-to-noise ratio conditions, the obtained Capon power spectrum and CFAR detector can be used to extract the distance and angle information of the target and interference of the h-th radar at the k-1th CPI:

[0068]

[0069] in, and They represent the real target measurement set and the false target measurement set respectively, n refers to the nth measurement, It represents the measurement number obtained by the h-th radar after CFAR detection at the k-1-th CPI.

[0070] Distance and angle can be modeled as nonlinear functions:

[0071]

[0072] Among them, x l and y l are the positions of the target in the X-axis and Y-axis directions in the Cartesian coordinate system, represents the nonlinear measurement function, To measure noise.

[0073] For H radars, the measurement set obtained at the same CPI can be written as

[0074] S23: Define the Euclidean distance matrix between any two radar measurements, which is expressed as:

[0075]

[0076] in, h and Represent the hth and radar, E is the Euclidean distance operator, represents the inverse operation of the nonlinear measurement function,

[0077] S24: A {0,1} indicator matrix is ​​established according to a pre-set distance threshold and a Euclidean distance matrix to identify main lobe deceptive trajectory interference and real targets, thereby obtaining a target recognition result.

[0078] Specifically, according to the pre-set distance threshold ζ and the Euclidean distance matrix Establish a {0,1} indicator matrix, where 1 represents a real target and vice versa.

[0079]

[0080] in,

[0081] According to the indicator matrix It can identify main lobe deceptive trajectory interference and real targets.

[0082] S3: Extract the target measurement information from the target recognition results of each radar respectively, and initialize the target state based on the G-pair large interval measurement drive.

[0083] Assume that after identification, within the k-1th CPI, the real target information observed by the H radar can be rewritten as:

[0084]

[0085] The measurement z of the real target for the hth radar h,k-1 , the corresponding Cartesian coordinates can be obtained using the measurement function (noised and with measurement errors):

[0086]

[0087] Therefore, for the hth radar, the measurement data of any two adjacent CPIs can be used to initialize the initial state of the target for the tracker to track the target.

[0088] However, due to detection errors, directly using data from two adjacent CPIs for state initialization can result in significant errors (particularly exacerbating velocity initialization errors). Therefore, this embodiment proposes a target state initialization algorithm based on G-pair large-interval measurement drive to mitigate state initialization errors caused by measurement errors. Taking the hth radar as an example, step S3 specifically includes:

[0089] S31: Set G-pair measurement, expressed as

[0090] in, and They represent the hth radar in the k-1th CPI and the k-1+Nth CPI respectively. I The g-th pair of measurements of the true target information observed within a CPI, 1≤g≤G;

[0091] S32: Initialize the target speed using G-pair measurement.

[0092] Specifically, the speed initialized using any pair of measurements is:

[0093]

[0094] Among them, x v,g and y v,g They represent the components of the target velocity, Represents the inverse operation of a nonlinear measurement function.

[0095] Then the speed of initialization using G-pair measurement is:

[0096]

[0097] S33: Initialize the target position using G-pair measurement;

[0098] Specifically, the target position initialized using the g-th pair of measurements is:

[0099]

[0100] Among them, x g and y g Represent the two components of the target position respectively.

[0101] Then the target position initialized by G-pair measurement is:

[0102]

[0103] It should be noted that the initialization target position and speed refer to the hth radar at the (k-1+N I ) CPI target state can be recorded as:

[0104]

[0105] This embodiment adopts the target state initialization method of G-pair large interval measurement drive to better ensure stable tracking.

[0106] S4: Use the EKF tracker and the initialized target state to predict the target state of each radar and update the corresponding target state.

[0107] S41. Utilize the EKF tracker and the initialized target state to predict the target state of each radar.

[0108] a) Predict the target state of the kth CPI of the hth radar:

[0109] If the hth radar is in the (k-1+N I ) CPI target state is initialized to For convenience, (k-1+N I ) is rewritten as (k-1), Rewritten as xh,k-1 , then the kth CPI target state can be predicted according to the state transition equation:

[0110] x h,k|k-1 =Fx h,k-1 +v

[0111] Among them, x h,k|k-1 represents the predicted target state, x h,k-1 represents the target state of the hth radar in the k-1th CPI, F is the linear state transfer matrix, and v is the process noise with mean 0 and covariance matrix Q, where:

[0112]

[0113] Among them, I n and 0 n is an n×n unit matrix and an all-0 matrix;

[0114]

[0115] Among them, α is used to control the size of the process noise covariance.

[0116] b) Predict the covariance matrix of the target state of the h-th radar at the k-th CPI:

[0117] Specifically, in this embodiment, in addition to state prediction, EKF also needs to predict the covariance matrix of the target state. Assume that the covariance matrix of the h-th radar at the k-1th CPI target state is initialized to P h,k-1 , then according to the EKF filter, the covariance matrix prediction of the k-th CPI target state is:

[0118] P h,k|k-1 =FP h,k-1 F T +Q

[0119] Among them, P h,k|k-1 represents the predicted covariance matrix, P h,k-1 represents the covariance matrix of the target state of the h-th radar within the k-1-th CPI.

[0120] S42. Calculate the optimal weight vector of the target state prediction result based on the maximum Capon power spectrum criterion, and use the optimal weight vector to suppress main lobe deceptive trajectory interference and extract true target measurement information.

[0121] Specifically, suppose the target state prediction value obtained by the EKF tracker of the hth mine at the kth CPI is x h,k|k-1 , then the predicted values ​​of distance and angle can be obtained according to the nonlinear measurement function {r h,k|k-1 ,θh,k|k-1}. {r h,k|k-1 ,θ h,k|k-1} can be used as the target prior information of the hth mine at the kth CPI. Assume r h,k|k-1 Located at nth bin range gates, then use {r h,k|k-1 ,θ h,k|k-1 The minimum variance distortion-free generator of this prior information is:

[0122]

[0123] Wherein, is the noise + interference covariance matrix, and its training samples are taken from all the samples except the (nth bin -N ε / 2) to (n bin +N ε / 2) range gates, N ε Indicates that the nth bin N range gates ε A distance gate. is the adaptive optimal weight calculated according to the Lagrange multiplier method, is the virtual steering vector, a(r h,k|k-1 ,θ h,k|k-1 ) and b(θ h,k|k-1 ) are the transmit steering vector and the receive steering vector respectively.

[0124] However, since the predicted target state x h,k|k-1 There must be a deviation between the actual target state, so it leads to There is an error between the virtual guidance vector and the real target, and this error will make the use of The main beam after beamforming deviates from the real target, so the EKF predicted value x is directly used h,k|k-1 Obtained It cannot be used directly for the most powerful solution.

[0125] To address this issue, this embodiment proposes an optimal weight solution method based on the maximum Capon power spectrum criterion:

[0126]

[0127] Among them, o∈{n bin -1,n bin ,n bin +1}. This is because the target state predicted by the EKF filter may be located at the range gate where the real target is located or at an adjacent range gate. r and Θ θDenote the feasible domain of distance and angle respectively. According to this optimization principle, the virtual guidance vector v(r,θ) of the k-th CPI real target can be obtained, and the minimum variance distortionless former (MVDR) can be rewritten as:

[0128]

[0129] Where w is the optimal weight vector obtained by solving the maximum Capon power spectrum criterion. Assume that and are located at the nth bin Range gate and nth θ angle units, then the nth bin Range gate n θ The power output formed by using the optimal weight w on each angle unit is:

[0130]

[0131] If the azimuth dimension is spaced by angle θ Δ Divided into N θ Angle units, the number of distance gates is N bin . Indicates that the hth radar is at the nth bin The kth CPI of a range gate includes the signal, interference and noise covariance matrices.

[0132] Traversing all range gates and angle units can obtain the filtered output on all range gates and angle units:

[0133]

[0134] At this time, since the main lobe deceptive jammer differs from the real target in the range dimension, it will be suppressed. Then the measurement value of the real target obtained by the h-th radar at the k-th CPI is:

[0135]

[0136] Among them, z h,k represents the measurement value of the real target obtained by the h-th radar at the k-th CPI, and They represent the target distance and angle information obtained by the h-th radar at the k-th CPI, and They represent the azimuth observation range and the distance observation range respectively.

[0137] This embodiment uses the optimal weight solution algorithm of the maximum Capon power spectrum criterion to suppress mainlobe deceptive trajectory interference and extract real target measurement information, which can ensure that the mainlobe of the radiation pattern always points to the distance and angle of the target, avoid the divergence of subsequent tracking, and improve the tracking accuracy.

[0138] S43. Update the target state of the EKF tracker according to the extracted real target measurement information.

[0139] Specifically, the target state and covariance matrix predicted by the EKF tracker are updated using the real target measurement information obtained by the h-th radar at the k-th CPI:

[0140]

[0141] P h,k =(IK h,k H h,k )P h,k|k-1

[0142] Among them, I is the unit matrix, K h,k is the Kalman gain:

[0143] K h,k =P h,k|k-1 H h,k T S h,k -1

[0144] S h,k It is new information The covariance of is the measurement function,

[0145] S h,k =H h,k P h,k|k-1 H h,k T +R h,k

[0146] is the Jacobian matrix, R h,k represents the measurement covariance matrix of the h-th mine at the k-th CPI, because the distance and angle information are obtained through searching, therefore,

[0147]

[0148] r Δ and θ Δ Respectively represent the range gate size and angle search interval, and Used to control the measurement noise amplitude.

[0149] It should be noted that after step S43, the following steps are also included:

[0150] S44. Correct the updated target state based on the target state correction method of the auxiliary particles.

[0151] Specifically, assume that after EKF update, the target state estimated by the h-th radar at the k-th CPI is Then use As the mean, P h,k,p As the sampling covariance, sampling N h,k,p particles that follow a Gaussian distribution

[0152] in, represents the sth particle, is the initialized particle weight, then,

[0153]

[0154] Define the mapping between target state and range gate number Then the particle weight can be updated as:

[0155]

[0156] in, Indicates the The covariance matrix of the signal on the range gate, the denominator is a normalization constant;

[0157]

[0158]

[0159] Then the target state after auxiliary particle correction is estimated to be:

[0160]

[0161] After obtaining the updated target state, this embodiment also uses the auxiliary particle-based target state correction algorithm to perform correction, effectively avoiding the tracking divergence problem caused by inaccurate initialization parameters of the traditional EKF tracker.

[0162] S5: Use the track fusion algorithm to fuse the target states obtained by each radar to obtain the final target state.

[0163] The target state and covariance set estimated by the H radar obtained by the EKF tracker is:

[0164]

[0165] The target state and covariance matrix can be converted to the same coordinate system and recorded as:

[0166]

[0167] in, It represents the target state after the h-th radar obtains the transformed coordinate system at the k-th CPI, and the covariance matrix P h,k The coordinate system remains unchanged after the coordinate system transformation. This is because the public coordinate system and the radar's own coordinate system are both in the xy two-dimensional plane, and the distribution characteristics of the target state do not change with the transformation of the coordinate system.

[0168] Assuming that the target state estimation errors of H radars are independent of each other, the target tracks generated by multiple radars can be fused using a simple track fusion algorithm. The target state covariance matrix after track fusion is:

[0169]

[0170] The target state after fusion is:

[0171]

[0172] The fused target state x k and the covariance matrix P k The coordinate systems of different radars are converted to estimate the target state at the next moment. After the coordinate conversion, the target state and covariance matrix of each radar are rewritten as:

[0173]

[0174] Because the target's true state is unknown, the tracking results of a single radar may have large errors. Track fusion can alleviate the tracking errors of a single radar. Providing the fused target states and covariances of each radar to the cognitive radar system can achieve sequential cognitive target recognition and tracking.

[0175] This embodiment improves the performance of a single FDA-MIMO radar with large target tracking error through a track fusion strategy, thereby improving the overall tracking effect.

[0176] S6: Repeat the operations of steps S4 and S5 to achieve closed-loop cognitive target tracking and output the target tracking trajectory.

[0177] The present invention first establishes the signal model of mainlobe deceptive trajectory interference and target, then identifies the target and mainlobe deceptive trajectory interference by utilizing FDA-MIMO radar networking, and finally uses the initialized target state for cognitive tracking. The method can accurately identify the real target and the mainlobe deceptive trajectory interference in the mainlobe deceptive trajectory interference environment, and the time delay between the interference and the target can be less than one pulse. In addition, the tracker is initialized when the target prior information is unknown, thereby realizing target tracking.

[0178] Example 2

[0179] Based on the first embodiment, this embodiment verifies and illustrates the multi-target tracking method of the cognitive FDA-MIMO radar network through simulation experiments.

[0180] 1. Experimental conditions:

[0181] Two radars are set up in the experiment, and the radar coordinates are [0m, 0m] and [-1500m, 0m]. The initial position of the target is [-1000m, 1000m], and the initial speed of the target is [180m / s, 20m / s]. During the experiment, three main lobe deceptive trajectory interferences are set for each radar. Their positions and speeds are generated according to the positions of the real target and the two radars to ensure that the angles of the interference and the target relative to the radar are consistent within different CPIs. The target signal-to-noise ratio is 10dB, and the interference-to-noise ratio is 30dB. The target state transition obeys uniform linear motion, and the process noise control coefficient α=2. The initial target state covariance matrix is ​​P=diag([50m 2 ,50m 2 ·s -2 ,50m 2 ,50m 2 ·s -2 ]).

[0182] The operating parameters of the two FDA-MIMO radars in the experiment are the same. The speed of light is set to c = 3 × 10 8 m / s, the number of transmitting and receiving array elements N = M = 12, and the carrier frequency f0 = 10 × 10 9 Hz, the frequency step of the signal transmitted by the transmitting element is Δf=25×10 3 Hz, wavelength λ=c / f0, spacing between array elements is d=λ / 2, pulse repetition frequency PRF=Δf, pulse repetition period PRT=1 / PRF, each CPI contains 500 pulses. Tracking step size T CPI =500×PRT. The target tracking duration is 50 CPI. The Capon power spectrum search range is [-π / 2,π / 2]×[0m,15×10 3 m]. The search intervals for angle and distance are r Δ =10m and θ Δ =0.1°, the measurement noise control coefficient is set to and For convenience, the backscatter coefficients of the target and interference are set to 1. The number of auxiliary particle samples is 2000, and the sampling covariance P = diag([50m 2 ,10m 2 ·s -2 ,50m 2 ,10m 2 ·s -2]).

[0183] 2. Experimental content and results analysis:

[0184] Under the above simulation parameters, the target tracking performance of the method of this embodiment is simulated, and the experimental results are analyzed.

[0185] See Figure 6 , Figure 6 The target tracking results of the cognitive FDA-MIMO radar network proposed in this paper are given. Among them, the first 10 CPIs are measured by real targets and interference, which are used for real target identification and track initialization of the cognitive FDA-MIMO radar network. The target tracking of the cognitive FDA-MIMO radar network is performed starting from the 11th CPI. Figure 6 From the measurements of Radar 1 and Radar 2, it can be seen that only true target measurements are obtained after the 10th CPI. This phenomenon demonstrates that the proposed cognitive FDA-MIMO radar network is able to identify and suppress mainlobe deceptive trajectory interference and obtain true target measurements.

[0186] From the above analysis, we can see that Figure 6 This shows that the cognitive FDA-MIMO radar network target recognition and tracking algorithm proposed in this invention can recognize the target and main lobe deception trajectory interference and output the target trajectory when the target prior information is completely unknown.

[0187] See Table 1, Figure 7-12 The algorithms used by Filter 1 to Filter 5 are shown in Table 1. Where '√' indicates that the corresponding algorithm is used in the filtering, and conversely, '×' indicates that the corresponding algorithm is not used in the filtering.

[0188] Table 1 Statistics of the usage of each algorithm

[0189]

[0190] See Figure 7 and Figure 8 , Figure 7 and Figure 8 are the RMSE of the target position estimation by Radar 1 and Radar 2, respectively. Comparing the position estimation results of Filter 1 and Filter 2, it can be seen that in the target tracking process of the cognitive FDA-MIMO radar network, the target state correction algorithm based on auxiliary particles proposed in this paper can effectively solve the tracking divergence problem caused by the initialization parameter errors (target position and velocity) of the traditional EKF filter.

[0191] Comparing Filter 2 and Filter 3, it can be seen that the estimation accuracy of the target position can be improved by adopting the optimal weight selection algorithm under the maximum power spectrum principle proposed in the present invention. Comparing Filter 3 and Filter 4, it can be seen that the cognitive FDA-MIMO radar network target tracking algorithm after trajectory fusion is more accurate in estimating the target position. From the position estimation result of Filter 5, it can be seen that if the G-pair large interval measurement driven target state initialization algorithm proposed in the present invention is not used to initialize the target state, the EKF tracker will diverge quickly and form an erroneous tracking result. On the contrary, if the G-pair large interval measurement driven target state initialization algorithm is used to initialize the target state, the target position can be estimated more effectively.

[0192] See Figure 9 and Figure 10 , Figure 9 and Figure 10 These are the RMSE of the target velocity estimated by Radar 1 and Radar 2, respectively. Figure 9 and Figure 10 In the example, Filter 1-Filter 4 all use the proposed G-pair large-interval measurement drive target state initialization algorithm, so Filter 1-Filter 4 can effectively estimate the target speed. In contrast, Filter 5 and Filter 6 will have a large initialization speed error. Figure 9 The reason why the velocity estimation error of Filter 5 gradually increases after the 11th CPI is that the initial velocity error of the target in Radar 2 is large, which indirectly leads to the decrease of the velocity estimation accuracy of Radar 1 after trajectory fusion. Figure 10 The reason why the target velocity estimation accuracy gradually decreases after Filter 5 initialization is that the velocity estimation accuracy of Radar 1 is low, so the velocity estimation accuracy of Radar 2 is improved after trajectory fusion.

[0193] See Figure 11 and Figure 12 ,Depend on Figure 11 and Figure 12It can be seen that among Radar 1 and Radar 2, Filter 1 achieves the maximum output signal-to-interference-and-noise ratio (SINR). This is because in Filter 1, the proposed optimal weight selection algorithm based on the maximum power spectrum principle, the G-pair large-interval measurement-driven target state initialization algorithm, the multi-radar trajectory fusion algorithm, and the auxiliary particle-based target state correction algorithm are all used. However, in Filters 2 through 5, as the auxiliary particle-based target state correction algorithm, the optimal weight selection algorithm based on the maximum power spectrum principle, the multi-radar trajectory fusion algorithm, and the G-pair large-interval measurement-driven target state initialization algorithm are successively eliminated, the output SINR gradually decreases. In particular, after the G-pair large-interval measurement-driven target state initialization algorithm is eliminated, the output SINR of Filter 5 also drops rapidly due to the rapid divergence of the target state estimate.

[0194] The above is a further detailed description of the present invention in conjunction with specific preferred embodiments, and the specific implementation of the present invention should not be considered to be limited to these descriptions. For those skilled in the art of the present invention, without departing from the concept of the present invention, several simple deductions or substitutions can be made, which should be considered to fall within the scope of protection of the present invention.

Claims

1. A target recognition and tracking method based on cognitive FDA-MIMO radar network, characterized in that: include: S1: Establish a measurement model for radar net targets and mainlobe deceptive trajectory jamming; S2: Based on the measurement model, the FDA-MIMO radar network is used to identify targets using mainlobe deceptive trajectory jamming and the spatial distribution characteristics of targets relative to different radars to obtain target recognition results for each radar; S3: Extract the target measurement information from the target recognition results of each radar, and initialize the target state based on the G-pair large-interval measurement drive; S4: Use the EKF tracker and the initialized target state to predict the target state of each radar and update the corresponding target state; S5: Use the track fusion algorithm to fuse the target states obtained by each radar to obtain the final target state; S6: Repeat the operations of steps S4 and S5 to achieve closed-loop cognitive target tracking and output the target tracking trajectory.

2. The target recognition and tracking method based on cognitive FDA-MIMO radar network according to claim 1 is characterized in that: In step S1, the measurement model of the radar net target and main lobe deceptive trajectory interference is expressed as: Where Y h,k represents the target measurement model of the hth radar in the kth CPI, Represents the target measurement of the hth radar within the kth CPI of the ith range gate, 1≤i≤N bin , N bin Indicates the number of range gates; represents the main lobe deceptive trajectory jamming measurement model of the kth CPI of the hth radar, and They are the nth generation of target and main lobe deceptive trajectory interference at the kth CPI after down-conversion matched filtering at the receiving end. bin The signal on the distance gate, All elements in the are subject to mean 0 and covariance σ h 2 The complex white Gaussian noise component; M h and N h They represent the number of transmitting array elements and receiving array elements of the h-th FDA-MIMO radar, and L represents the number of pulses accumulated within a coherent integration time.

3. The target recognition and tracking method based on cognitive FDA-MIMO radar network according to claim 1 is characterized in that: Step S2 includes: S21: Calculate the number of each radar at the nth bin Capon power spectrum within the k-1th CPI of a range gate; S22: Using the obtained Capon power spectrum and CFAR detector, the distance and angle information of the target and interference of each radar at the k-1th CPI are extracted to obtain the measurement set of all radars; S23: Set the Euclidean distance matrix between the measurements of any two radars; S24: A {0,1} indicator matrix is ​​established according to a pre-set distance threshold and a Euclidean distance matrix to identify main lobe deceptive trajectory interference and real targets, thereby obtaining a target recognition result.

4. The target recognition and tracking method based on cognitive FDA-MIMO radar network according to claim 1 is characterized in that: In step S3, initializing the target state based on the G-pair large interval measurement drive includes: S31: Set G-pair measurement, expressed as in, and They represent the hth radar in the k-1th CPI and the k-1+Nth CPI respectively. I The g-th pair of measurements of the true target information observed within a CPI, 1≤g≤G; S32: Initialize the target speed using G-pair measurement, which is expressed as: in, x v,g and y v,g They represent the components of the target velocity, Represents the inverse operation of the nonlinear measurement function, T CPI Indicates the length of a coherent accumulation time; S33: Initialize the target position using G-pair measurement, which is expressed as: in, x g and y g Represent the two components of the target position respectively.

5. The target recognition and tracking method based on cognitive FDA-MIMO radar network according to claim 1 is characterized in that: Step S4 includes: S41, using the EKF tracker and the initialized target state to predict the target state of each radar; S42, solving the optimal weight vector of the target state prediction result based on the maximum Capon power spectrum criterion, and using the optimal weight vector to suppress the main lobe deceptive trajectory interference and extract the real target measurement information; S43. Update the target state of the EKF tracker according to the extracted real target measurement information.

6. The target recognition and tracking method based on cognitive FDA-MIMO radar network according to claim 5 is characterized in that: Step S41 includes: a) Predict the target state of the kth CPI of the hth radar according to the following formula: x h,k|k-1 =Fx h,k-1 +v Among them, x h,k|k-1 represents the predicted target state, x h,k-1 represents the target state of the hth radar in the k-1th CPI, F is the linear state transfer matrix, and v is the process noise with mean 0 and covariance matrix Q; Among them, I n and 0 n is an n×n unit matrix and an all-zero matrix, α is a coefficient used to control the covariance of process noise, T CPI Indicates the length of a coherent accumulation time; b) Predict the covariance matrix of the target state of the h-th radar at the k-th CPI according to the following formula: P h,k|k-1 =FP h,k-1 F T +Q Among them, P h,k|k-1 represents the predicted covariance matrix, P h,k-1 represents the covariance matrix of the target state of the h-th radar within the k-1-th CPI.

7. The target recognition and tracking method based on cognitive FDA-MIMO radar network according to claim 6, characterized in that: In step S42, the formula for suppressing main lobe deceptive trajectory interference and extracting true target measurement information using the optimal weight vector is: Among them, z h,k represents the measurement value of the real target obtained by the h-th radar at the k-th CPI, and They represent the target distance and angle information obtained by the h-th radar at the k-th CPI, and Respectively represent the azimuth observation range and the distance observation range; N bin Indicates the number of range gates, N θ represents the number of angle units, w is the optimal weight vector obtained by solving the maximum Capon power spectrum criterion, H represents the transpose, Indicates that the hth radar is at the nth bin The kth CPI of a range gate includes the signal, interference and noise covariance matrices.

8. The target recognition and tracking method based on cognitive FDA-MIMO radar network according to claim 7, characterized in that: Step S43 includes: The target state and covariance matrix predicted by the EKF tracker are updated using the real target measurement information obtained by the h-th radar at the k-th CPI. The update formula is as follows: P h,k =(I-K h,k H h,k )P h,k|k-1 in, Indicates the updated target state, represents the nonlinear measurement function, K h,k is the Kalman gain, P h,k represents the updated covariance matrix, I is the unit matrix, H h,k represents the Jacobian matrix after the measurement function is approximately linearized.

9. The target recognition and tracking method based on cognitive FDA-MIMO radar network according to claim 5, characterized in that: After step S43, the method further includes: S44. Correct the updated target state based on the target state correction method of the auxiliary particles.

10. The target recognition and tracking method based on cognitive FDA-MIMO radar network according to claim 9, characterized in that: Step S44 includes: The updated target state of the hth radar at the kth CPI is As the mean, P h,k,p As the sampling covariance, sampling N h,k,p particles that obey a Gaussian distribution in, represents the sth particle, is the initialized particle weight, then, Define the mapping between target state and range gate number Then the particle weight can be updated as: in, Indicates the The covariance matrix of the signal on the range gate, the denominator is a normalization constant; Then the target state after auxiliary particle correction is estimated to be:

Images