Real-time networking radar system beam dwell scheduling method based on time pointer vector

Abstract

The present invention belongs to the field of radar system resource management, and particularly relates to a method for adaptive beam dwell scheduling of a distributed radar networking system. The present invention first introduces a time pointer vector in the dwell scheduling analysis of the radar networking system to improve the time utilization rate of the system. Considering the importance and urgency criteria in dwell scheduling, the present invention selects the task with the highest comprehensive priority at the current scheduling analysis moment, and this comprehensive priority takes into account both the working mode priority and the deadline of the task. To ensure that the finally obtained actual execution task queue can meet the expected execution time criterion, for the selected task above, among all the radar sets that can execute this task, the radar node with the smallest time offset rate is selected to execute this task. In addition, the present invention can achieve real-time beam dwell scheduling analysis, and thus can be applied to actual radar systems.

Description

Technical Field

[0001] The present invention belongs to the field of radar system resource management, and particularly relates to a method for adaptive beam dwell scheduling of a networked radar system. Background Art

[0002] Compared with a single-station phased array radar, a distributed multi-functional radar networking system composed of multiple radar nodes has better detection, tracking, parameter estimation, interference suppression, and anti-fading capabilities (see the literature: Zhou Wenhui. Research on Resource Management Technology of Phased Array Radar and Networking Tracking System [D]. National University of Defense Technology, 2004.). To maximize the performance of the distributed radar networking system, an efficient real-time beam dwell scheduling algorithm needs to be designed accordingly.

[0003] The existing research on radar beam dwell scheduling algorithms mainly focuses on single-radar systems. The literature (Lu Jianbin, Hu Weidong, Yu Wenxian. Research on real-time task scheduling of multi-functional phased array radars [J]. Acta Electronica Sinica, 2006, 34(4): 732-736.) proposed a beam dwell scheduling algorithm based on time pointers for phased array radars. In this algorithm, the deadline and working mode priority of tasks are comprehensively considered to form dynamic priorities, and the task with the highest comprehensive priority is scheduled at each analysis moment. The literature (Zhang H, Xie J, Zong B, et al. Dynamic priority scheduling method for the air-defence phased array radar [J]. IET Radar Sonar & Navigation, 2017, 11(7): 1140-1146.) considered the threat degree of targets in the design of dynamic priorities. Based on this, an online beam dwell scheduling algorithm was designed to obtain the corresponding actual scheduling sequence. The literature (Qu Z, Ding Z, Moo P. Dual-side scheduling for radar resource management [C]. International Radar Symposium, Warsaw, Poland, 2020: 260-263.) proposed a dual-side beam dwell scheduling algorithm. This algorithm sets a separation point in the scheduling analysis time period, and schedules the beam dwell tasks on both sides of the separation point according to the earliest deadline principle and the principle that the start time is closest to the separation point respectively. The literature (Chen Y, Zhang Q, Yuan N, et.al. An adaptive ISAR-imaging-considered task scheduling algorithm for multi-function phased array radars [J]. IEEE Transactions on Signal Processing, 2015, 63(19): 5096–5110.) designed the priority of imaging tasks and implemented the scheduling of radar search, tracking, and imaging tasks with a heuristic scheduling method. In the literature (Duan Yi, Tan Xiansi, Qu Zhiguo, et al. Phased array radar event scheduling method based on variable time window [J]. Modern Radar, 2018, 40(2): 1-6.), the concept of variable time window was introduced, and a phased array radar beam dwell scheduling algorithm based on variable time window was designed, which further increased the adjustment range of the actual execution moment of tracking tasks and improved the scheduling success rate of the system.The literature (Yang S, Tian K, Liu R. Task scheduling algorithm based on value optimization for anti-missile phased array radar[J]. IET Radar Sonar & Navigation, 2019, 13(11): 1883–1889.) proposed a beam dwell scheduling model based on value optimization, and used a genetic algorithm to obtain the optimal solution to this problem and the corresponding optimal scheduling sequence. For the beam dwell scheduling problem under supersonic target tracking, the literature (Meng F, Tian K. Phased-array radar task scheduling method for hypersonic-glide vehicles[J]. IEEE Access, 2020, 8: 221288-221298.) solved this problem using an algorithm based on particle swarm-simulated annealing.

[0004] Different from the beam dwell scheduling under a single-station radar, the beam dwell scheduling under a distributed radar networking system needs to determine not only the actual execution time of each dwell task but also the matching relationship between each task and radar nodes. The literature (Shaghaghi M, Adve R S. Task selection and scheduling in multifunction multichannel radars[C]. IEEE Radar Conference, Seattle, USA, 2017:0969-0974.) extended the Earliest Deadline First (EDF) algorithm, which is used to solve the multi-pipeline scheduling problem, to the beam dwell scheduling under multi-channel radars, and simultaneously proposed a beam dwell scheduling method under multi-channel radars based on the Branch & Bound (B&B) method. The literature (Shaghaghi M, Adve R S. Machine learning based cognitive radar resource management[C]. IEEE Radar Conference, Oklahoma City, USA, 2018:1433-1438; Shaghaghi M, Adve R S, Ding Z. multifunction cognitive radar task scheduling using monte carlo tree search and policy networks[J]. IET Radar Sonar&Navigation., vol.12, no.12, pp.1437–1447, 2018; Xu L, Zhang T. Reinforcement learning based dynamic task scheduling for multifunction radar network[C], IEEE Radar Conference, Florence, Italy, 2020:1-5.) respectively implemented the beam dwell scheduling under a distributed system using machine learning, Monte Carlo tree search, and Q-learning.The literature (Liu X, Zhang Q, Luo Y, et al. ISAR imaging task allocation for multi-target in radar network based on potential game[J]. IEEE Sensor Journal, 2019, 19(23): 11192-11204.) adopted a beam dwell scheduling method based on convex optimization to achieve beam dwell in a distributed system considering imaging tasks.

[0005] Certain achievements have been made in beam dwelling under the above-mentioned distributed system, but there are still the following problems: (1) The methods proposed in the above-mentioned literature do not consider the expected time criterion for beam dwelling scheduling, which means that the actual execution time of a task should be as close as possible to its expected execution time; (2) Although the beam dwelling scheduling algorithm using the heuristic method (see the literature: Shaghaghi M, Adve R S. Task selection and scheduling in multifunction multichannel radars[C]. IEEE Radar Conference, Seattle, USA, 2017:0969-0974.) has a small computational complexity and is real-time, the performance of the obtained scheduling sequence is not good and cannot meet the importance and urgency criteria for beam dwelling scheduling. And using intelligent methods (see the literature: Shaghaghi M, Adve R S. Task selection and scheduling in multifunction multichannel radars[C]. IEEE Radar Conference, Seattle, USA, 2017:0969-0974; Shaghaghi M, Adve R S. Machine learning based cognitive radar resource management[C]. IEEE Radar Conference, Oklahoma City, USA, 2018:1433-1438; Shaghaghi M, Adve R S, Ding Z. multifunction cognitive radar task scheduling using monte carlo tree search and policy networks[J]. IET Radar Sonar&Navigation., vol.12, no.12, pp.1437–1447, 2018; Xu L, Zhang T. Reinforcement learning based dynamic task scheduling for multifunction radar network[C], IEEE Radar Conference, Florence, Italy, 2020:1-5; Shaghaghi M, Adve R S.The scheduling method in [C]. IEEE Radar Conference, Oklahoma City, USA, 2018: 1433 - 1438. (Machine learning based cognitive radar resource management) is not real - time and cannot be applied to the actual beam dwell scheduling analysis.

[0006] Based on the above problems, the present invention proposes a beam dwell scheduling method for a real - time networking radar system based on a time pointer vector. By introducing the time pointer vector, the present invention improves the time utilization rate of the distributed radar networking system, and schedules the beam dwell task with the highest comprehensive priority at the moment indicated by the time pointer vector to the node with the lowest time offset rate, so that the algorithm can effectively take into account the importance, urgency and expected execution time criteria of beam dwell scheduling. Summary of the Invention

[0007] The present invention proposes a beam dwell scheduling method for a real - time networking radar system based on a time pointer vector, which is characterized in that:

[0008] Assume that within the current scheduling interval [t0, t0 + t SI , there are N dwell tasks T = [T1, T2,..., T N applying for scheduling on M phased - array radars R = [R1, R2,..., R M , where t0 is the start time of the current scheduling interval, t SI is the duration of the scheduling interval, and (t0 + t SI ) is the end time of the current scheduling interval. The dwell task model is T i = {W i dt i l i dw i Pt i}, where W i is the working mode priority, dt i is the expected execution time, l i is the time window, dw i is the dwell duration, and Pt i is the transmit power. The beam dwell scheduling method for a real - time networking radar system based on a time pointer vector includes the following steps:

[0009] Step 1: Initialize the time pointer vector tp = [tp1, tp2,..., tp M = [t0, t0,..., t0] corresponding to each radar;

[0010] Step 2: Select the tasks in the task request queue T that satisfy dt i +l i <min(tp), delete them from it, and store them in the task deletion queue;

[0011] Step 3: Select all the tasks in the task request queue T that satisfy max(tp)≥dt i -l i . Suppose there are X tasks selected. If X > 0, calculate the priority sw of each task according to the following formula i :

[0012]

[0013] where Xd i is the serial number of the task request T i (1≤i≤X) arranged from largest to smallest deadline among the X tasks, and Xp i is the serial number arranged from smallest to largest working mode priority among the X tasks. Sort these X tasks from largest to smallest according to the comprehensive priority, initialize itp = 1 and go to Step 4; if X = 0, then go to Step 8;

[0014] Step 4: Take the itp-th task T in the sorted task queue itp , select the radar set R itp that can be used to schedule T itp , and the radar R itp in R j should meet the following requirements:

[0015] dt i -l i ≤tp j ≤dt i +l i ∩tp j +dw i ≤t0+t SI ,R j ∈R itp (2)

[0016] Step 5: If R itp is not empty, then go to Step 6; if R itp is empty and itp < X, then itp = itp + 1, and return to Step 4; if R itp is empty and itp = X, then go to Step 8;

[0017] Step 6: Schedule T itp to the radar j* in R itp that makes the time offset rate of T itp the smallest:

[0018]

[0019] where tp itp is the set of time pointers corresponding to the radar in R itp , and 1 is a vector of all 1s with the same dimension as R itp ;

[0020] Step 7: Let the actual execution time of T itp be tp j* , put T itp into the task execution queue of R j* , and delete T itp from the task request queue, update tp j * = tp j* + dw itp and directly go to Step 9;

[0021] Step 8: Update min(tp) = min(tp) + Δt, where Δt is the minimum sliding step of the time pointer;

[0022] Step 9: If T is not empty and min(tp) < t0 + t SI , then return to Step 2, otherwise the analysis of this scheduling interval ends.

[0023] Principle of the Invention

[0024] The radar beam dwell scheduling needs to follow the importance criterion, the urgency criterion and the expected execution time criterion. The importance criterion indicates that the radar system should schedule as many high operating mode priority tasks as possible. The urgency criterion indicates that tasks with earlier deadlines should be given priority as much as possible. And the expected execution time criterion indicates that the actual execution time of each task should be as close as possible to the expected execution time. According to these three criteria of the dwell scheduling, the following scheduling benefit function is constructed for each beam dwell task:

[0025] G i (dt i , at i , l i , W i , t0) = g1(W i )g2(dt i , l i , t0)g3(dt i , at i , l i ) (4)

[0026] where at i is the actual execution time of the task, g1(W i ), g2(dt i , l i , t0) and g3(dti , at i , l i ) The calculation formula is as follows:

[0027] g1(W i ) = W i (5)

[0028]

[0029]

[0030] Since g1(W i ) increases as the priority of the task working mode increases, this term reflects the importance criterion of scheduling; in g2(dt i , l i , t0), c1 is a normal constant. Since it increases as the deadline of the task decreases, this term reflects the urgency criterion of scheduling. In g3(dt i , at i , l i ), c2 is also a normal constant. Since it increases as the difference |dt i - at i | between the actual execution time and the expected execution time of the task decreases, this term reflects the expected execution time criterion of the task. Assume that within the current scheduling interval [t0, t0 + t SI , there are N resident tasks T = [T1, T2,..., T N applying for scheduling on M phased array radars R = [R1, R2,..., R M . Combining the above scheduling revenue function and the constraint conditions in the distributed system scheduling problem, the mathematical model of the distributed system beam dwell scheduling problem can be established as follows:

[0031]

[0032] Among them, represents the number of tasks actually executed on radar R p . N2 and N3 are the number of delayed tasks and the number of deleted tasks respectively. N1 is the sum of the total number of all actually executed tasks, that is N2 and N3 should satisfy the first constraint condition. The second constraint condition means that all scheduled tasks must be executed within their executable time range. The third constraint condition means that the actual execution times of different resident tasks on the same radar cannot overlap. The fourth and fifth inequalities represent the conditions that the delayed and deleted tasks should satisfy. The above problem is a typical NP-hard problem.

[0033] To maximize the objective function in (8), considering that the objective function requires scheduling important and more urgent tasks, the task with the highest comprehensive priority at the current scheduling analysis moment is selected. This comprehensive priority takes into account both the working mode priority and the deadline of the task, as shown in Steps 3 and 4, to determine the task for the current scheduling analysis. Further, the objective function also considers the expected execution time criterion. Therefore, for the above-selected task, among all the radar sets that can execute this task, the radar node with the minimum time offset rate is selected to execute the task, as shown in Step 6. In summary, the present invention schedules the beam application dwell task with the maximum comprehensive priority at the moment indicated by the time pointer to the radar node with the minimum time offset rate at the current moment, thereby satisfying the three criteria of beam dwell scheduling and obtaining an actual execution task queue that meets the beam dwell scheduling constraint conditions and has a high scheduling gain. Description of the Drawings

[0034] Figure 1 TDR comparison of three methods

[0035] Figure 2 HVR comparison of three methods

[0036] Figure 3 TUR comparison of three methods

[0037] Figure 4 ATSR comparison of three methods

[0038] Figure 5 Running time comparison of three methods Detailed Implementation Manner

[0039] In the simulation scenario, five types of tasks are considered, namely precision tracking, ordinary tracking, horizon search, airspace search, and verification. Among them, the horizon search task has three areas to search, and the airspace search task also has three areas to search. The ratio of the number of precision tracking targets to the number of ordinary task targets is 1:4. The radar task parameters are shown in Table 1. It is assumed that there are a total of three phased array radars in the distributed radar networking system, and each radar can execute all the dwell tasks. The simulation duration is 12 s, the scheduling interval duration is set to 50 ms, and Δt = 0.5 ms.

[0040] Table 1. Radar Beam Dwell Task Parameter Table

[0041]

[0042] To comprehensively evaluate the performance of the present invention, in this section, the Task Drop Ratio (TDR), Hit Value Ratio (HVR), Time Utilization Ratio (TUR), Average Time Shifting Ratio (ATSR), and running duration are used as performance evaluation metrics. The above metrics are defined as follows:

[0043] The Task Drop Ratio (TDR) is the ratio of the number of lost tasks to the number of tasks applied for scheduling within the simulation duration:

[0044] TDR = N drop / N all (9)

[0045] Where N drop represents the number of lost tasks, and N all represents the number of tasks applied for scheduling;

[0046] The Hit Value Ratio (HVR) is the ratio of the sum of the working mode priorities of the actually executed tasks to the sum of the working mode priorities of the tasks applied for scheduling within the simulation duration:

[0047]

[0048] Where N exe represents the number of actually executed tasks. This metric is used to reflect the proportion of high-priority tasks successfully scheduled;

[0049] The Time Utilization Ratio (TUR): It is defined as the ratio of the total residence duration of the actually executed tasks to the product of the total simulation duration and the number of radar nodes:

[0050]

[0051] Where t total is the total simulation duration, and M is the total number of radar nodes in the distributed networking system.

[0052] The Average Time Shifting Ratio (ATSR): It is defined as the average of the time shifting ratios of each executed tracking task:

[0053]

[0054] Where N tra is the total number of actually executed tracking tasks.

[0055] The beam dwell scheduling method of the real-time networking radar system based on the time pointer vector proposed in the present invention (referred to as the invention method in the simulation) is used for performance comparison with Method A and Method B. Method A is a networking radar beam dwell scheduling algorithm based on the earliest deadline first algorithm (see the literature: Shaghaghi M, Adve R S. Task selection and scheduling in multifunction multichannel radars[C]. IEEE Radar Conference, Seattle, USA, 2017: 0969-0974.), and Method B is the algorithm obtained by extending the algorithm proposed in the literature (Lu Jianbin, Hu Weidong, Yu Wenxian. Research on real-time task scheduling of multifunction phased array radars[J]. Acta Electronica Sinica, 2006, 34(4): 732-736.) to the networking radar beam dwell scheduling. This algorithm schedules the task with the highest comprehensive priority to the radar node with the smallest time pointer. The simulation platform is MATLAB R2019a, the computer processor is Core i7-10700, and the memory is 16G. Figures 1 to 5 It is the statistical result of 100 Monte Carlo simulations under different metrics.

[0056] Figure 1 It is the task loss rate curve. When the number of targets is 10, obvious task losses begin to occur in Method A. This is because the earliest deadline first algorithm does not introduce a time pointer, resulting in the failure of each dwell task to be closely arranged on the time axis, thus leading to premature task losses. While both Method B and the invention method introduce a time pointer during the scheduling analysis, ensuring that there are no time gaps in the actual execution queue of tasks and improving the utilization rate of time resources. Therefore, their task loss rate curves are lower than that of Method A.

[0057] Figure 2 It is the achieved value rate curve, which reflects whether tasks with higher working mode priorities are executed as many as possible. This curve has an opposite trend to the loss rate curve. It can be seen that both the present method and Method B have greatly improved the achieved value rate.

[0058] Figure 3 It is the time utilization rate curve. When the number of targets is small, as the number of targets increases, the time utilization rate of the invention method in the present invention increases linearly. When the number of targets reaches about 60, the networking system reaches a saturation state, and the trend of this curve tends to be stable. It can be seen that the time utilization rates of both the invention method and Method B are higher than that of Method A.

[0059] Figure 4It is the average time offset rate curve. Since the inventive method takes into account the expected time criterion in the scheduling process, it can be seen that the inventive method has the lowest average time offset rate compared with the other two methods.

[0060] Figure 5 It is the average running duration of the three methods in a scheduling interval. It can be seen that Method A has the lowest average running duration, while for Method B and the inventive method, since a time pointer is introduced in the dwell scheduling analysis, the algorithm complexity is increased, resulting in the running time of these two algorithms being higher than that of Method A. There is almost no difference in the running duration between Method B and the inventive method, which indicates that considering the expected time criterion hardly brings additional computational complexity to the inventive method. The average running duration of the three methods does not exceed one scheduling interval, so all three algorithms have real-time performance.

[0061] In summary, compared with Method A, the beam dwell scheduling method for the network radar system proposed by the present invention has a lower task loss rate, a higher implementation value rate and time utilization rate; compared with Method B, the inventive method effectively reduces the time offset rate; in addition, the inventive method has real-time performance and can be applied to an actual radar system.

Claims

1. A beam dwell scheduling method for a real-time networking radar system based on a time pointer vector, characterized by: Assume that within the current scheduling interval [t0, t0 + t SI , there are N resident tasks T = [T1, T2,..., T N applying for scheduling on M phased array radars R = [R1, R2,..., R M , where t0 is the start time of the current scheduling interval, t SI is the duration of the scheduling interval, and (t0 + t SI ) is the end time of the current scheduling interval; the resident task model is T i = {W i dt i l i dw i Pt i}, where W i is the working mode priority, dt i is the expected execution time, l i is the time window, dw i is the dwell duration, and Pt i is the transmit power; the real-time networking radar system beam dwell scheduling method based on the time pointer vector includes the following steps: Step 1: Initialize the time pointer vector tp = [tp1, tp2,..., tp M = [t0, t0,..., t0]; Step 2: Select the tasks in the task request queue T that satisfy dt i +l i <min(tp), delete them from the queue, and store them in the task deletion queue; Step 3: Select all tasks in the task request queue T that satisfy max(tp)≥dt i -l i Assume there are X tasks selected. If X > 0, calculate the priority sw of each task according to the following formula i : Among them, Xd i is the task request T i (1 ≤ i ≤ X) is the serial number arranged in descending order of deadlines among X tasks, and Xp i is the serial number arranged in ascending order of working mode priorities among X tasks. Sort these X tasks in descending order of comprehensive priorities, initialize itp = 1 and go to step 4; if X = 0, then go to step 8; Step 4: Take the itp-th task T in the sorted task queue itp , and select the set of radars R that can be used to schedule T itp . The radars R itp in R itp should meet the following requirements: j ​ dt i -l i ≤tp j ≤dt i +l i ∩tp j +dw i ≤t0+t SI ,R j ∈R itp (2) Step 5: If R itp is not empty, go to Step 6; if R itp is empty and itp < X, then itp = itp + 1, and return to Step 4; if R itp is empty and itp = X, then go to Step 8; Step 6: Schedule T itp to the radar j* in R itp such that the time offset rate of T itp is minimized: Among them, tp itp is R itp the set of time pointers corresponding to the radar in R, and 1 is a vector of all 1s with the same dimension as R itp vector; Step 7: Let the actual execution time of T itp be tp j* , put T itp into the task execution queue of R j* , and delete T itp from the task request queue, update tp j* = tp j* + dw itp and directly go to Step 9; Step 8: Update min(tp) = min(tp) + Δt, where Δt is the minimum sliding step of the time pointer; Step 9: If T is not empty and min(tp) < t0 + t SI , then return to Step 2; otherwise, the analysis of this scheduling interval ends.

Images