Abstract
In order to save the radar resources and obtain the better low probability intercept ability in the network, a novel radar selection method for target tracking based on improved interacting multiple model information filtering (IMM-IF) is presented. Firstly, the relationship model between radar resource and tracking accuracy is built, and the IMM-IF method is presented. Then, the information gain of every radar is predicted according to the IMM-IF, and the radars with larger information gain are selected to track target. Finally, the weight parameters for the tracking fusion are designed after the error covariance prediction of every working radar, in order to improve the IMM-IF. Simulation results show that the proposed algorithm not only saves much more radar resources than other methods but also has excellent tracking accuracy.
1. Introduction
Multiple radar systems have been shown to offer significant advantages over traditional monostatic radars [1]. When it comes to the reasonable configuration of multiple radar systems, the concept of resource scheduling becomes very important and has been drawing more and more attention in recent years.
As we know, low probability intercept (LPI) is one of the important features of modern radars. LPI optimization strategy is proposed in radar network architectures in [2, 3], where transmit power is minimized among netted phased array radars. The paper [4] studies the problem of scheduling the searching, verification, and tracking tasks of the military surveillance radar. The continuous double auction parameter selection algorithm is presented in [5] to solve the problem of allocating resource and selecting operational parameters for phased array radar system. Networked phased array radars that are connected by a communication channel are studied in paper [6], which proposes two types of distributed management techniques for the coordinated radar resource management. Based on sparsity-aware matrix decomposition, an improved method to select informative radars for target tracking in radar networks in proposed in [7], which shows the efficiency and improvement of the tracking performance. The paper [8] presents an optimal solution to the power allocation problem in distributed active multiple-radar systems subject to different power constraints, using a linear fusion rule and a simple objective function. In paper [9], a novel general criterion with the consideration of correlation in the measurement errors from different radars is proposed, which is applicable for a more general sensor network with an excellent tracking performance.
Sensor data fusion techniques are widely used in target tracking of radar network, which can be loosely defined as how to best extract useful information from multiple radar observations. Information filtering, which is essentially a Kalman filter expressed in terms of the inverse of the covariance matrix, has been widely used in multiple sensor estimation [10]. A novel approach to track-to-track fusion in a high-level sensor data fusion architecture for automotive surround environment perception using information matrix fusion (IMF) is presented in [11]. The paper [12] utilizes the iterative joint-integrated probabilistic data association technique for multisensor distributed fusion systems. The paper [13] develops the square-root extensions of unscented information filter and central difference information filter, which have better numerical properties than the original versions. For nonlinear system, the paper [14] presents a new state estimation algorithm called square-root cubature information filter, which is further extended for use in multisensory state estimation. To achieve further improvements in tracking accuracy, a self-adaptive algorithm for maneuver strategy transition probability matrix is proposed in [15], by utilizing the compression ratio of the maneuver strategy error. Many different architectures are presented in [16] for the target tracking in 3D Cartesian coordinates, with the measurements of all the sensors in polar coordinates.
However, those papers focus on the data selection after radar detection and do not consider the difference and accuracy of the sensors. Less working radars means much better LPI performance. In order to save the radar resource with excellent LPI performance, a radar selection algorithm is proposed based on an improved information filtering. The remainder of this paper is organized as follows. Section 2 presents the improved information filtering method and radar selection method. Simulations of the proposed algorithms and comparison results with other methods are provided in Section 3. The conclusions are presented in Section 4.
2. Radar Selection Based on Improved Information Filter
Interacting multiple model (IMM) method is used for tracking maneuvering target. Information filter (IF), which is the dual Kalman filter, has attracted much attention for tracking fusion using multiple sensors [17]. But it cannot present excellent tracking performance for maneuvering target. In order to accomplish tracking fusion of maneuvering targets in radar network, a novel radar selection method based on interacting multiple model information filtering (IMMIF) is proposed in this section.
2.1. Tracking Models and Measurement Noise in Radar Network
All the dynamic models are , is the th model used at time , the switch probability from model to model is , , , . is the probability of model , . Let and represent the state vector and the observation vector, respectively; the state equation and transfer equation at time are where and are stationary white noise processes with covariance matrices and . is the transition matrix and is the observation matrix. Every recurrence of the IMM algorithm contains interacting of input, model’s filtering, update of model probability, and interacting of output.
The covariance matrix of measurement noise is controlled by the emitted power. As we know, radar equation at time is as follows: where is the single dwelling time of the beam from the normal direction at time , is the average radiated power, is the receiver gain, is the radar cross section (RCS) of the target, is Boltzmann constant, and are, respectively, effective noise temperature and radar system loss, is the detection range, is the transmit gain, represents the signal to noise ratio of the system at time .
The single pulse signal is radiated by the radar, and the covariance of the measurement noise can be denoted as: where is the pulse width, is the wave velocity, and is the carrier frequency. We can see that different can lead to different .
2.2. Information Filtering for Every Model
Utilizing all the states and model probabilities from last recurrence, the computation of input state and covariance of model can be express as
Using information state and Fisher information replace state estimate and covariance , we can obtain information filtering result based on Kalman filter. The definition of information state and Fisher information is
Prediction and estimation of and and can be obtained by recursive iteration, combining with and . The prediction of information state and Fisher information are given as
In the light of observation data from different sensors in the sensor network, the estimation of information state and Fisher information for every model are as follows: where and are the contributions that measurements make to and , respectively, which can be represented as follows.
The model probability is recursively updated by the th radar as where is the likelihood function of model at time .
In the radar network, every radar will update a model probability, the final model probability can be calculated as
2.3. Emitted Radar Selection Based on Information Computation
According to formula (2) and (3), the covariance of the measurement noise can be predicted based on the prediction of target distance and target RCS. So in (2) is replaced by which is predicted by and . is presented as
and are the target’s range and velocity which are estimated by the IMM tracking algorithm at time and is the tracking interval. RCS is supposed to be the same as , which has been measured by the fusion center according to the echo.
There are radars in the network. When the observation noise is predicted by the th radar, the information can be calculated using where is the information gain which can be predicted by (9).
An information vector can be represented as
is the mean value of , which can be represented as
There are radars whose information gain are larger than , which will be selected as the working radars at next time, .
2.4. Tracking Fusion Based on the Prediction for Error Covariance Matrix
The selected radars are used for target tracking. Then, the observation vector will be obtained. Using information filter for every model, the final estimation of the th radar can be represented as where .
Different radars have different predicted covariance matrix. The predicted covariance matrix is given as where is model probability at time , the predicted covariance matrix of every model can be represented as .
Then estimations and predicted covariance matrices and its trace will be obtained, TRn = trace . Then, the reciprocal vector is computed. The weight of the radar tracking data can be formulated as
The final fusion result can be obtained at last:
3. Simulation Results
In this section, Monte Carlo simulations are performed to analyze the performance of the proposed resource scheduling method.
3.1. Trajectory Design
Figure 1 shows the target trajectory with its measurement results in 100 s. RCS of every radar is produced randomly during target tracking. All the radar positions are shown in Figure 1, which can be used to evaluate the LPI performance of the fusion methods.
3.2. Radar Selection Results
To verify the effectiveness of the proposed approach based on the detailed mathematical expressions of Section 2, computer simulations using MATLAB software are conducted for the comparisons with other methods [14–16]. Measurement data fusion [16] is used in all the methods in the simulations. The radiation label of the five radars is shown in Figure 2. We can see that the radars work in turn. The number of working numbers at every tracking time is illustrated in Figure 3; we can see that the number of working numbers is less than 5 at most times. Compared with the traditional fusion methods [14–16], which use all the sensors or radars for the fusion, we can see that the proposed method reduces much more working radars. As a result, the proposed method has better low probability intercept ability than others with less radiation time. In the proposed strategy, the radars can be selected to work according to targets’ positions and radar information gain in order to meet the requirement of desired tracking performance.
3.3. Comparison of Tracking Performance
The proposed adaptive radar selection method is labeled as “adaptive fusion.” Traditional fusion methods using all the radars and using selected radars based on Section 2.3 are labeled as “all radar-based traditional fusion” and “selected radar-based traditional fusion”, respectively. The root-mean-square error (RMSE) of time can be formulated as (20) where is the number of the Monte Carlo simulation, is the true state of the system, and is the estimated vector at the th simulation, .
Figures 4 and 5 show the range RMSE using the three methods. We can see that the proposed method “adaptive fusion” presents best tracking accuracy with other methods.
From the simulation results, we can see the proposed radar selection strategy is effective, which can achieve an optimal trade-off between the working radar numbers and tracking accuracy. The proposed algorithm selects the radars which can provide more accurate tracking data and then improves the information filtering by designing the weight parameters which can realize the tracking fusion perfectly. But “all radar-based traditional fusion” uses more radars for target tracking without tracking data selection, as some radars obtain the tracking data with much measurement noises or process noises, which will deteriorate tracking fusion performance. And “selected radar-based traditional fusion” does not consider the validity of the radar tracking data, and the fusion is realized using average weight of every radar. In addition, the common information filtering method does not consider the contribution difference of the radars in the network.
4. Conclusions
In this paper, we have presented a new radar selection method for the radar network based on the improved IMM information filtering method. During the target tracking, the relation model is built between the radar resource and tracking performance. Then the radars are selected by the predicted information gain, and the weight parameters for fusion are designed in order to obtain the best tracking accuracy. The simulation results show that the proposed algorithm reduces much more radars with excellent tracking performance.
Data Availability
The data used to support the findings of this study are available from the corresponding author upon request.
Conflicts of Interest
The authors declare that there is no conflict of interest regarding the publication of this paper.
Acknowledgments
The authors would like to thank the anonymous reviewers for their insightful comments and suggestions that have contributed to improve this paper. We note that there are no data sharing issues since all of the numerical information is produced by solving the equations of the proposed algorithm, which are realized by MATLAB software in the paper. This work was supported by the National Natural Science Fund (61871203, 61401179) in China, China Postdoctoral Science Foundation (2016M592334), and Qing Lan Project of Jiangsu Province.