Simultaneous multi-beam radar beam residence real-time scheduling method based on double time pointers

Abstract

The invention belongs to the field of radar system resource management, and particularly relates to a real-time adaptive residence scheduling method of a simultaneous multi-beam radar system. The method comprises the following steps: firstly, setting a separation point in a scheduling interval, and respectively adopting a wave beam residence scheduling method based on time pointers when tasks are executed on the left side and the right side of the separation point to form a scheduling mode based on the double time pointers; then, under the constraint of the total aperture resource, time, frequency and electromagnetic compatibility of the system, task analysis with the maximum comprehensive priority is selected at each time pointer; in the comprehensive priority design, a task working mode priority is combined with a deadline or an earliest executable moment to construct a first-level comprehensive priority, and then a second-level comprehensive priority is constructed together with a deviation between an expected execution moment and an actual execution moment. When the aperture resources at the time pointers are insufficient or the tasks meeting the constraint conditions are completely analyzed, the time pointers are updated to the left side or the right side, the updated step length is the maximum residence duration in the multiple tasks executed at the same time, and the process is repeated till the time pointers on the two sides move to the end point of the scheduling interval. And finally, selecting a task scheduling sequence with the minimum cost as a final scheduling result from the wave beam resident scheduling sequences obtained under the condition of different separation points. Simulation results show that compared with an existing method, the method gives consideration to priorities, deadlines and expected execution time criteria, well balances the loss rate and the time migration degree, and has real-time performance.

Description

Technical Field

[0001] The invention belongs to the field of radar system resource management, and in particular relates to a method for real-time adaptive resident scheduling of a simultaneous multi-beam radar system. Background Art

[0002] In an increasingly complex war environment, phased array radar (PAR) is widely used due to its advantages such as multi-function and multiple working modes, which largely relies on the simultaneous multi-beam capability of phased array radar. Phased array radar can emit multiple independent beams at the same time, so it can perform tasks such as simultaneous multi-target monitoring, tracking, positioning, and imaging. At present, the methods for realizing multi-beam are as follows: (1) utilizing the beam agility characteristics of phased array radar to send beams with different directions within the pulse repetition period, thus approximately realizing simultaneous multi-beam (Chen J, Tian Z, Wang L, et al. Adaptive simultaneous multi-beam dwell scheduling algorithm for multifunction phased array radars [J]. Journal of Information & Computational Science, 2011, 8 (14): 3051-3061.); (2) utilizing the phased array aperture to divide into several sub-arrays, each of which can transmit a beam independently, thus realizing simultaneous multi-beam (Xue G, Du Z, Wei W, et al. Multi-beam dwell adaptive scheduling algorithm for helicopter-borne radar [C] Information Technology & Artificial Intelligence Conference. 2014: 401-404.). In order to give full play to the effectiveness of the simultaneous multi-beam radar system, an effective beam dwell scheduling method needs to be designed.

[0003] For the design of phased array radar beam dwell scheduling strategy, the existing methods focus on adaptive scheduling algorithms. The literature (Lu Jianbin, Hu Weidong, Yu Wenxian. Research on real-time task scheduling of multifunctional phased array radar [J]. Journal of Electronics, 2006, 34 (4): 5.) proposed an adaptive beam dwell scheduling algorithm, in which the concept of time pointer is introduced to enable the task priority to change dynamically during the scheduling process. The priority takes into account the two parameters of the task's working mode priority and deadline. 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.) further considers the threat level of the target in the setting of the comprehensive priority. However, the above method fails to take into account the expected execution time criterion, which requires that the actual execution time of the dwell task should be as close as possible to the expected execution time of the task. Based on this problem, the paper (Cheng T, He Z, Tang T. Dwell scheduling algorithm for multifunction phased array radars based on the scheduling gain [J]. Systems Engineering and Electronics: English Edition, 2008, 19 (3): 7.) proposed a beam dwell scheduling method based on scheduling interval analysis. This method sorts tasks according to the task scheduling gain and assigns actual execution time to each dwell task in turn according to the expected time criterion. This method can effectively take into account the expected execution time criterion, but it has problems such as low time utilization and poor real-time performance. The paper (Qu Z, Ding Z, Moo P. Dual-Side Scheduling for Radar Resource Management [C] 21st International Radar Symposium (IRS), 2020: 260-263) proposed a dual-side scheduling method, which sets a separation point in a scheduling interval, divides a scheduling interval into two sides, and schedules tasks from the separation point to both sides of the scheduling interval. Although this method can reduce the deviation between the actual execution time and the expected execution time, it does not consider the deadline and work mode priority during the scheduling process, and the loss rate is high.The literature (Lu H, Yang X, Hu S. Task scheduling of improved time shifting based on genetic algorithm for phased array radar [C] 13th International Conference on Signal Processing (ICSP). IEEE, 2017.) designed a fitness function based on the importance, urgency and expected execution time of the task, and used a genetic algorithm to obtain the optimal actual execution task queue. The literature (Shaghaghi M, Adve RS, Zhen D. Multifunction Cognitive Radar Task Scheduling Using Monte Carlo Tree Search and Policy Networks [J]. IET Radar, Sonar & Navigation, 2018, 12 (12): 1437-1447.) designed an objective function for the multifunctional cognitive beam dwell scheduling problem based on the urgency and expected time criteria of the task, and used the Monte Carlo search tree optimization method to obtain the optimal actual execution task sequence and its actual execution time. Although the above two beam dwell scheduling methods based on optimization algorithms take into account the expected execution time criteria, their real-time performance is poor, so they cannot be used in actual radar beam dwell scheduling.

[0004] Although the above-mentioned literature has achieved some results in radar beam dwell scheduling, they all focus on single-beam systems. The literature (Chen J, Tian Z, Wang L, et al. Adaptive simultaneous multi-beam dwell scheduling algorithm for multifunction phased array radars [J]. Journal of Information & Computational Science, 2011, 8 (14): 3051-3061.) proposed an adaptive simultaneous multi-beam dwell scheduling method. However, this method does not essentially achieve the simultaneous transmission of multiple beams. It only uses pulse overlap technology to transmit beams in different directions within a pulse repetition period, thereby improving the time utilization of the system. The literature (Xue G, Du Z, Wei W, et al. Multi-beam dwell adaptive scheduling algorithm for helicopter-borne radar [C] Information Technology & Artificial Intelligence Conference. 2014: 401-404.) uses aperture sharing to achieve simultaneous multi-beam dwell scheduling. However, its scheduling algorithm is based on the analysis method of scheduling interval, so it generates multiple time gaps, resulting in low time utilization. Reference (Chen Cheng, Ma Xiaoyan, Yang Ruijuan, Cheng Wei. Research on task scheduling of multifunctional electronic system based on aperture segmentation [J]. Journal of Air Force Radar Academy, 2012, 26(06): 412-418.) Based on the analysis method based on scheduling interval, electromagnetic compatibility constraints are further considered in the problem model. Although the reference (Qi Wenchao, Yang Ruijuan, Li Xiaobai, Chen Xinyong, Cheng Wei. Research on task scheduling algorithm of multifunctional integrated radar [J]. Radar Science and Technology, 2012, 10(02): 150-155.) proposes a scheduling method for simultaneous multi-beam system based on time pointer to improve time utilization, it only considers time and energy resource constraints, and the set constraints are not comprehensive. The literature (Zhu Xitong, Yang Ruijuan, Li Xiaobai, Yuan Kai. Resource scheduling method of multifunctional integrated system based on improved GA-PSO [J]. Journal of Air Force Early Warning Academy, 2021, 35(03): 202-206+220.) proposed an improved genetic-particle swarm algorithm. In the problem model, the complete constraints were considered, but the calculation time was long and it was not real-time.

[0005] Based on the above problems, the present invention proposes a real-time scheduling method for beam dwelling of a simultaneous multi-beam radar based on dual time pointers. The method is based on the idea of ​​setting a separation point in the literature (Qu Z, Ding Z, Moo P. Dual-Side Scheduling for Radar Resource Management [C] 21st International Radar Symposium (IRS), 2020: 260-263) and introduces dual time pointers. In addition, for the beam dwell scheduling of a simultaneous multi-beam system, the method proposed in the present invention takes into account comprehensive constraints. When executing tasks on the left and right sides of the separation point, the beam dwell scheduling method based on time pointers is respectively adopted, forming a scheduling method based on dual time pointers. Subsequently, under the constraints of the system's total aperture resources, time, frequency and electromagnetic compatibility, the task analysis with the largest comprehensive priority is selected at each time pointer, and the comprehensive priority considers the three criteria for scheduling. When the aperture resources at the time pointer are insufficient or the task analysis that meets the constraints is completed, the time pointer is updated to the left or right, and the step size of the update is the maximum dwell time among several tasks executed simultaneously, and the above process is repeated until the time pointers on both sides move to the end point of the scheduling interval. Finally, among the beam dwell scheduling sequences obtained under different separation points, the task scheduling sequence with the minimum cost is selected as the final scheduling result. The simulation results show that compared with the existing methods, this method takes into account the priority, deadline and expected execution time criteria, better balances the loss rate and time deviation, and has real-time performance. Summary of the invention

[0006] Assume that in the current scheduling interval t0,t end There are N resident tasks T = T1, T2, ..., T N Apply for scheduling, where t0 is the start time of the current scheduling interval, t end is the end time of the current scheduling interval. The resident task model is T i =rt i ,st i ,l i ,p i ,t i ,A i ,β i ,v i , where rt i is the expected execution time, st i is the actual execution time, l i is the time window, p i is the working mode priority, Δt i is the residence time, A i Indicates the aperture resource requirement of the task. Discretize the frequency domain and spatial domain resources and number them, βi T i Occupied frequency band number, v i T i The simultaneous multi-beam radar beam dwell real-time scheduling method based on dual time pointers includes the following steps:

[0007] Step 1: In a scheduling interval, evenly select L points, denoted as s1:s L , let j=1.

[0008] Step 2: Let s = s j ,tp1=s j ,tp2=s j ,, from s j Click Start to select task scheduling from the left and right. i (i=1,...,N) successively judge whether it belongs to s j The left task or the right task, specifically, if T i The expected execution time satisfies formula (1), then it is s j Click the task on the left, otherwise it is s j The task on the right side of the point, that is, the side to which the task's residence time belongs entirely or mostly, is the side to which the task belongs. Suppose there are N1 tasks on the left side and N2 tasks on the right side.

[0009] |rt i -s j |>|rt i +Δt i -s j | (1)

[0010] Step 3: Analyze the tasks on the left.

[0011] Step 3.1: If the left task satisfies tp1-Δt i <rt i -l i , then store these tasks in the deletion queue dele, record n i Tasks are deleted, i=i+n i ,η1=A total , η1 represents the remaining aperture resources on the left.

[0012] Step 3.2: In the task request queue, assume that the left task satisfies t0≤tp1-Δt i ≤(rt i +l i ) has X1 tasks. If X1=0, then tp1=tp1-Δtp min , Δtp minis the minimum sliding step of the time pointer, and proceed to step 3.8. If X1>0, proceed to the next step.

[0013] Step 3.3: The first-level integrated priority is shown in formula (2). When executing the left task, the time pointer tp1 moves to the left, Xe i (1≤i≤X1) is the sorting sequence number of the X1 tasks from the earliest execution time to the largest, Xp i It is the sequence number in which the working mode priorities among X1 tasks are arranged from small to large.

[0014]

[0015] Then, the first-level integrated priority is combined with the time offset to construct the second-level integrated priority:

[0016]

[0017] Among them, X 1i It is task T i In X1 tasks, according to the first-level comprehensive priority sw 1i The sequence number is sorted from small to large, Xs i For task T i In X1 tasks, according to the time offset The order from large to small. ω determines the weight of the time offset and the first-level comprehensive priority, and the value range is [1, X1+1]. The tasks are sorted from large to small according to the second-level comprehensive priority, and itp1=1.

[0018] Step 3.4: Take the itp1th task in the sorted task queue

[0019] Step 3.5: If Judgement i Is there a conflict with the task whose actual execution time is tp1? If there is no conflict, proceed to the next step; otherwise, continue to judge There is a conflict with the task whose actual execution time is tp1. If there is no conflict, proceed to the next step; otherwise, determine whether there is an idle frequency band available. If so, allocate an idle frequency band to it, and then proceed to the next step; if not or Then proceed to step 3.8.

[0020] Step 3.6: Put it into the execution queue Sche(j) and Delete from the task request queue, i=i+1, The residence time is stored in t_store, Otherwise, go directly to the next step.

[0021] Step 3.7: itp1=itp1+1. If itp1≤X1 and η1>0, return to step 3.4, otherwise proceed to the next step.

[0022] Step 3.8: After the scheduling analysis at time tp1 is completed, update the time pointer tp1 = tp1-max(t_store), and set η1 = A total .

[0023] Step 3.9: If tp1≤t0 or i=N1, the analysis on the left side of this scheduling interval ends and proceeds to step 10, otherwise return to step 3.

[0024] Step 4: Analyze the task on the right side and set i=0.

[0025] Step 4.1: If the right-hand task satisfies tp2>rt i +l i , then store these tasks in the deletion queue dele, record n i Tasks are deleted, i=i+n i ,η2=A total , η2 represents the remaining aperture resources on the right.

[0026] Step 4.2: Assume that rt is satisfied for the right-hand task i -l i ≤tp2≤(t end -Δt i ) has X2 tasks. If X2=0, then tp2=tp2+Δtp min , go to step 4.8. If X2>0, go to the next step.

[0027] Step 4.3: Use the following formula to calculate the first-level comprehensive priority sw i :

[0028]

[0029] Among them, Xd i Request T for the task i (1≤i≤X2) The number of tasks in X2 arranged in descending order according to their deadlines, Xp i It is the sequence number of the X2 tasks arranged in ascending order according to the working mode priority. Then the second-level comprehensive priority is constructed by combining the first-level comprehensive priority and the time offset:

[0030]

[0031] Among them, X 1i It is the first-level comprehensive priority sw among X2 tasks 1i The sequence number is sorted from small to large, Xs iShift the time for X2 tasks Sort the tasks by their secondary comprehensive priority from large to small, and set itp2=1.

[0032] Step 4.4: Take the itp2th task in the sorted task queue

[0033] Step 4.5: If A itp2 ≤η2, judge v i Is there a conflict with the task whose actual execution time is tp2? If there is no conflict, proceed to the next step; otherwise, continue to judge β itp2 There is a conflict with the task whose actual execution time is tp2. If there is no conflict, proceed to the next step; otherwise, determine whether there is an idle frequency band available. If so, allocate an idle frequency band to it, and then proceed to the next step; if not or A itp2 >η2, then go to step 4.8.

[0034] Step 4.6: Put it into the execution queue Sche(j) and Delete from the task request queue, i=i+1, The residence time is stored in t_store, η2 = η2-A itp2 , otherwise go directly to the next step.

[0035] Step 4.7: itp2=itp2+1. If itp2≤X2 and η2>0, return to step 4.4, otherwise proceed to the next step.

[0036] Step 4.8: The scheduling analysis at time tp2 is completed, and the time pointer tp2 = tp2 + max(t_store) is updated, and η2 = A total .

[0037] Step 4.9: If tp2 ≥ t end Or i=N2, the analysis on the right side of this scheduling interval ends.

[0038] Step 5: Set s=s j The scheduling sequence Sche(j) at that time is stored, and the cost Cos(j) of selecting Sche(j) as the final scheduling sequence is calculated using formula (6), where the cost Cos(i,j) of task i is calculated using formula (7), j=j+1.

[0039]

[0040]

[0041] C dpis the penalty factor for task loss, and d describes the impact of priority on the cost of task loss.

[0042] Step 6: If j>L, then set s for the current scheduling interval The corresponding scheduling sequence is the optimal scheduling sequence, and the scheduling interval analysis ends. Otherwise, return to step 2.

[0043] Principle of the Invention

[0044] The radar beam dwell scheduling process needs to follow three criteria, including importance, urgency, and expected execution time. According to the three scheduling criteria, the following scheduling benefit function is constructed for each beam dwell scheduling task:

[0045] G i (st i ,rt i ,l i ,p i ,t0,t end )=g1(p i )g2(rt i ,l i ,t0,t end )g3(st i ,rt i ,l i ) (8)

[0046] in,

[0047]

[0048] Because g1(p i ) increases with the priority of the task working mode, so this item reflects the importance of scheduling; g2(rt i ,l i ,t0,t end ) where c1 is a positive constant, which increases as the deadline of the task decreases, so this term reflects the urgency criterion of scheduling; g3(st i ,rt i ,l i ) where c2 is a positive constant, since g3(st i ,rt i ,l i ) decreases as the deviation between the actual execution time and the expected execution time increases, so this item reflects the expected execution time criterion of scheduling. Considering the objective function and constraints in the beam dwell scheduling problem, the mathematical model of the beam dwell scheduling problem is established as follows:

[0049]

[0050] Where N1, N2 and N3 are the number of scheduled tasks, the number of delayed tasks and the number of deleted tasks respectively. Obviously, N = N1 + N2 + N3. In the optimization model, A i It is task T i The aperture resource demand of I is the set of tasks with overlapping execution time periods. ij For task T i With T j Whether the frequency occupancy overlaps, if the frequency occupancy overlaps, it is 1, otherwise it is 0; V ij Indicates whether there is overlap in spatial occupancy, if the same is 1, otherwise it is 0. In the optimization model, the first inequality indicates that the task must be executed within its scheduling time range; the second and third inequalities reflect the conditions that delayed tasks and deleted tasks should meet; the fourth inequality indicates that for tasks with overlapping execution time periods, the sum of the aperture resources used does not exceed the total system aperture resources; the fifth inequality indicates that for tasks with overlapping execution time periods, the sum of the frequency resources used does not exceed the total system frequency resources; the last inequality indicates that tasks with overlapping execution time periods cannot overlap in both the frequency domain and the spatial domain.

[0051] Although the application of the traditional time pointer analysis method and the scheduling interval-based method in the simultaneous multi-beam system can effectively solve the above problem model, the former selects the task scheduling execution with higher working mode priority and earlier deadline at each analysis moment, so it cannot consider the expected execution time criterion, while the latter will produce many time gaps, so the time utilization rate is low. The literature (Qu Z, Ding Z, Moo P. Dual-Side Scheduling for Radar Resource Management [C] 21st International Radar Symposium (IRS), 2020: 260-263) aims at the problem that the EST (the earliest start time) beam residence scheduling algorithm cannot consider the expected execution time, sets multiple separation points to implement multiple scheduling, and selects the scheduling sequence that minimizes the cost function. Inspired by this algorithm, the present invention considers combining the time pointer-based beam residence scheduling method with this idea, so as to reduce the time offset while ensuring the time utilization rate. In step 1, L separation points are set. Due to the introduction of separation points, the tasks applying for scheduling execution are divided into two parts, that is, in step 2, the tasks are divided into the left and right sides, so there are left and right scheduling on both sides of the separation point. In order to maximize time utilization and schedule as many high-priority tasks as possible, the beam dwell scheduling method based on time pointer is adopted when scheduling on both sides of the separation point, and then the dual time pointer is introduced. When scheduling, the importance criterion and the urgency criterion respectively require that tasks with high working mode priority and tasks with late earliest execution time (left task sequence) or early deadline (right task sequence) should be executed as much as possible. In steps 3.3 and 4.3, equations (2) and (4) reflect the above two criteria. Equations (3) and (5) introduce time offset into the comprehensive priority, reflecting the expected execution time criterion. Steps 3.5 and 4.5 take into account aperture resources, frequency and electromagnetic compatibility constraints. First, it is determined whether the aperture resource constraint is met. If it is met, the spatial domain and frequency domain conflict judgment is performed. If both exist at the same time, the electromagnetic compatibility constraint is not met. At this time, the occupied frequency band of the task needs to be adjusted. If the frequency resource constraint is met, that is, there are remaining frequency resources, then the occupied frequency band of the task currently being analyzed can be adjusted. Otherwise, this task cannot be successfully executed and the analysis of the current time pointer ends. Within a scheduling interval, when calculating the cost of the task scheduling sequence obtained at each separation point in step 5, formula (7) comprehensively considers the impact of time offset and lost tasks, and sets a penalty factor when the task is lost. While reducing the time offset, the task loss rate is reduced. When constructing the cost function, we believe that task loss is more important than time offset, and the time offset value range is [0,1], so the penalty factor is set greater than 1.Since the search task does not contain the expected execution time, no time offset will be generated. Therefore, in formula (7), the time offset of the executed search task is 0. From formulas (6) and (7), it can be seen that the fewer tasks are lost, the smaller the time offset and the smaller the cost. Therefore, in step 6, after traversing L separation points, according to the cost of the task sequence at each separation point, the task sequence with the smallest cost is selected as the final task scheduling result. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 The original task sequence and the scheduling sequence of the three methods

[0053] Figure 2 There are three methods of TDR;

[0054] Figure 3 There are three methods of HVR;

[0055] Figure 4 It is the TUR of three methods;

[0056] Figure 5 It is the ATSR of three methods;

[0057] Figure 6 is the running time of the three methods DETAILED DESCRIPTION

[0058] Four tasks are considered in the simulation scenario: precision tracking, general tracking, horizon search, and airspace search. The simulation duration is 12 seconds, the scheduling interval is set to 50 ms, and the ratio of the number of precision tracking targets to the number of general task targets is 1:4. In step 1, L is set to 6 to ensure that at least one task can be executed on the left and right sides of the separation point within a scheduling interval. In step 5, C dp Set to 10 to ensure that the cost of losing a task is greater than the time offset, and the time offset range is [0,1]. Set d to 1. When determining the final task scheduling sequence, when the number of lost tasks is the same, the impact of priority on the cost of task loss is greater than the time offset, so d ≥ 1. After simulation experiments, it is found that C dp When / d=10,d=[1,10], the task loss rate and the average time deviation rate can be better balanced. (when dispatching from the left) or (When scheduling on the right) it can better balance the time offset and loss rate. For the electromagnetic compatibility constraints in the problem model, the airspace and frequency domain are discretized and numbered, and a total of 5 airspaces and 4 frequency bands are set in the simulation. Each task occupies up to 3 airspaces, and the number and number of airspaces occupied by tasks are randomly generated. If the airspaces occupied by two tasks have the same number, there is an airspace conflict; similarly, each task is initially assigned a default frequency band, and the same number of two tasks indicates a frequency domain conflict. The task parameters are shown in Table 1:

[0059] Table 1. Radar beam dwell mission parameters

[0060]

[0061] In order to comprehensively evaluate the performance of the present invention, this section uses Task Drop Ratio (TDR), Hit Value Ratio (HVR), Time Utilization Ratio, Average Time Shifting Ratio (ATSR) and runtime as performance evaluation indicators. The above indicators are defined as follows:

[0062] The task loss rate (TDR) is the ratio of the number of lost tasks to the number of scheduled tasks during the simulation time:

[0063] TDR=N drop / N total (11)

[0064] Among them, N drop Indicates the number of lost tasks, N total Indicates the number of tasks requested for scheduling;

[0065] The realized value rate (HVR) is the ratio of the sum of the work mode priorities of the tasks actually executed during the simulation time to the sum of the work mode priorities of the tasks applied for scheduling:

[0066]

[0067] Among them, N exe Indicates the number of tasks actually executed. This indicator is used to reflect the proportion of high-priority tasks that are successfully scheduled;

[0068] Time utilization ratio (TUR): It is defined as the ratio of the total residence time of the actual execution task to the total simulation time:

[0069]

[0070] Among them, t total is the total simulation time.

[0071] Average Time Deviation Rate (ATSR): reflects the average deviation rate between the actual execution time and the expected execution time of all tracking tasks:

[0072]

[0073] Among them, N tra Indicates the number of tracking tasks actually executed.

[0074] Runtime: The runtime within a scheduling interval.

[0075] The real-time phased array radar beam dwell scheduling method based on dual time pointers proposed in the present invention is used for beam dwell scheduling, and the performance is compared with method A and method B. Method A is to apply the phased array radar beam dwell scheduling method based on time pointer (see document: Lu Jianbin, Hu Weidong, Yu Wenxian.: 'Research on real-time task scheduling of multifunctional phased array radar', Journal of Electronics, 2006, 34, (4), pp.732-736) to a simultaneous multi-beam system, and method B is a dwell scheduling method for a simultaneous multi-beam system based on a scheduling interval analysis method proposed in the document (Xue G, Du Z, Wei W, et al. Multi-beam dwell adaptive scheduling algorithm for helicopter-borne radar [C] Information Technology & Artificial Intelligence Conference. 2014: 401-404.). The simulation platform is MATLAB R2019a, the computer processor is Core i7-10700, and the memory is 16G.

[0076] Figure 1 The original sequence of a certain scheduling interval and the sequences after scheduling by three methods show the scheduling characteristics of the three methods. (a) is the original sequence, in which there are many overlapping tasks. (b) is the scheduling sequence of the method of the present invention. At this time, the separation point is selected at the position of 40ms, so the time pointer moves to both sides. (c) is the scheduling sequence of method A, and the time pointer moves from the start time to the end time of the scheduling interval. (d) is the scheduling sequence of method B. Since it is a scheduling method that selects the actual execution time for the task, there are many time gaps in the scheduling sequence. Through the scheduling of the three methods, aperture resource sharing is utilized to determine the task timing, which effectively solves the overlap between tasks.

[0077] Figures 2 to 6 The statistical results of 100 Monte Carlo tests under different indicators.

[0078] Figure 2is the task loss rate curve. When the number of targets increases to 60, the three methods begin to experience obvious task loss. As the number of targets increases, method B has a lot of time gaps and low time utilization, so it has the highest loss rate. Method A introduces a time pointer for scheduling, so that the tasks within the scheduling interval can be arranged tightly, so it has a low task loss rate. When scheduling tasks to both sides of the separation point, the present invention adopts the time pointer method in method A, and traverses the separation points at different positions to obtain the scheduling sequence with the lowest cost, so the task loss rate of the present invention is reduced.

[0079] Figure 3 The realization value rate curve reflects whether the tasks with higher priority in the working mode are executed as much as possible. The realization value rate of the method of the present invention is the highest when the target number is 0-90. Although it is slightly lower than method A when the target number is 100, the loss rate is higher than method A. This is because the loss rate of the method of the present invention for low-priority tasks is slightly lower than that of method B, and the loss rate of high-priority tasks is slightly higher than that of method B. The realization value rate of method B is the lowest, which is opposite to the trend of its loss rate curve.

[0080] Figure 4 is the time utilization curve. When the number of targets is small, the time utilization curve increases linearly with the increase of the number of targets; when the number of targets increases to a certain extent, tasks are lost and the time utilization of the system tends to be saturated. The method proposed in the present invention has the maximum time utilization.

[0081] Figure 5 is the average time offset rate curve. When the number of targets is 0, there is no tracking task, so there is no time offset, and the curve starts when the number of targets is 10. Method A has the highest time offset rate, and the method of the present invention is between the two. This is because method B arranges tasks as close to their expected execution time as possible during scheduling, while method A does not consider the impact of time offset. The method of the present invention takes into account the impact of time offset based on the design of the comprehensive priority and cost function of the task, thereby reducing the time offset rate compared with method A.

[0082] Figure 6 is the average running time of the three methods in one scheduling interval. The method of the present invention needs to traverse the separation points; when determining the actual execution time for the task, method B needs to find the time point closest to the expected execution time that meets the constraints within the task time window, so the calculation speed of the two methods is slow. However, the operation time of the three methods is within 50ms, which is real-time.

[0083] In summary, compared with method A, the method proposed in the present invention can reduce the task loss rate while reducing the time offset rate of the task; compared with method B, the present method greatly reduces the task loss rate and has a higher realization value rate, and the present method ensures the real-time nature of the scheduling analysis and can be applied to actual radar systems. The present invention is a real-time beam dwell scheduling algorithm that takes into account the priority, deadline and expected execution time criteria at the same time.

Claims

1. A real-time scheduling method for simultaneous multi-beam radar beam dwelling based on dual time pointers, characterized by: Assume that in the current scheduling interval t0,t end There are N resident tasks T = T1, T2, ..., T N Application scheduling, where t0 is the start time of the current scheduling interval, t end is the end time of the current scheduling interval; the resident task model is T i =rt i ,st i ,l i ,p i ,Δt i ,A i ,β i ,v i , where rt i is the expected execution time, st i is the actual execution time, l i is the time window, p i is the working mode priority, Δt i is the residence time, A i Indicates the aperture resource requirement of the task; discretizes the frequency domain and spatial domain resources and numbers them, β i T i Occupied frequency band number, v i T i The occupied airspace sequence number; the simultaneous multi-beam radar beam dwelling real-time scheduling method based on dual time pointers comprises the following steps: Step 1: In a scheduling interval, evenly select L points, denoted as s1:s L , let j = 1; Step 2: Let s = s j ,tp1=s j ,tp2=s j ,, from s j Click Start to select task scheduling to the left and right; for T i (i=1,...,N) successively judge whether it belongs to s j The left task or the right task, specifically, if T i The expected execution time satisfies formula (1), then it is s j Click the task on the left, otherwise it is s j Click the task on the right, that is, the task's residence time belongs entirely or mostly to which side, then this task belongs to that side; suppose there are N1 tasks on the left and N2 tasks on the right; |rt i -s j |>|rt i +Δt i -s j | (1) Step 3: Analyze the left task; Step 3.1: If the left task satisfies tp1-Δt i <rt i -l i , then store these tasks in the deletion queue dele, record n i Tasks are deleted, i=i+n i ,η1=A total , η1 represents the remaining aperture resources on the left; Step 3.2: In the task request queue, assume that the left task satisfies t0≤tp1-Δt i ≤(rt i +l i ) has X1 tasks; if X1 = 0, then tp1 = tp1 - Δtp min , Δtp min is the minimum sliding step of the time pointer, and proceed to step 3.8; if X1>0, proceed to the next step; Step 3.3: The first-level integrated priority is shown in formula (2); when executing the left task, the time pointer tp1 moves to the left, Xe i (1≤i≤X1) is the sorting sequence number of the X1 tasks from the earliest execution time to the largest, Xp i It is the order of priority of working modes in X1 tasks from small to large; Then, the first-level integrated priority is combined with the time offset to construct the second-level integrated priority: Among them, X 1i It is task T i In X1 tasks, according to the first-level comprehensive priority sw 1i The sequence number is sorted from small to large, Xs i For task T i In X1 tasks by time offset The order from large to small; η determines the weight of time offset and first-level comprehensive priority, and the value range is [1, X1+1]. Sort the tasks from large to small according to the second-level comprehensive priority, and let itp1=1; Step 3.4: Take the itp1th task in the sorted task queue Step 3.5: If Judgement i Is there a conflict with the task whose actual execution time is tp1? If there is no conflict, proceed to the next step; otherwise, continue to judge There is a conflict with the task whose actual execution time is tp1. If there is no conflict, proceed to the next step; otherwise, determine whether there is an idle frequency band available. If so, allocate an idle frequency band to it, and then proceed to the next step; if not or Then proceed to step 3.

8. Step 3.6: Put it into the execution queue Sche(j) and Delete from the task request queue, i=i+1, The residence time is stored in t_store, Otherwise, go directly to the next step; Step 3.7: itp1=itp1+1. If itp1≤X1 and η1>0, return to step 3.4, otherwise proceed to the next step; Step 3.8: After the scheduling analysis at time tp1 is completed, update the time pointer tp1 = tp1-max(t_store), and set η1 = A total ; Step 3.9: If tp1≤t0 or i=N1, the analysis on the left side of this scheduling interval ends and proceeds to step 10, otherwise return to step 3. Step 4: Analyze the task on the right side, let i = 0; Step 4.1: If the right-hand task satisfies tp2>rt i +l i , then store these tasks in the deletion queue dele, record n i Tasks are deleted, i=i+n i ,η2=A total , η2 represents the remaining aperture resources on the right side; Step 4.2: Assume that rt is satisfied for the right-hand task i -l i ≤tp2≤(t end -Δt i ) has X2 tasks; if X2=0, then tp2=tp2+Δtp min , go to step 4.8; if X2>0, go to the next step; Step 4.3: Use the following formula to calculate the first-level comprehensive priority sw i : Among them, Xd i Request T for the task i (1≤i≤X2) The number of tasks in X2 arranged in descending order according to their deadlines, Xp i The order of the X2 tasks in ascending order of working mode priority is then combined with the first-level integrated priority and the time offset to construct the second-level integrated priority: Among them, X 1i It is the first-level comprehensive priority sw among X2 tasks 1i The sequence number is sorted from small to large, Xs i Shift the time for X2 tasks Sort the tasks by their secondary comprehensive priority from large to small, and set itp2 = 1; Step 4.4: Take the itp2th task in the sorted task queue Step 4.5: If A itp2 ≤η2, judge v i Is there a conflict with the task whose actual execution time is tp2? If there is no conflict, proceed to the next step; otherwise, continue to judge β itp2 There is a conflict with the task whose actual execution time is tp2. If there is no conflict, proceed to the next step; otherwise, determine whether there is an idle frequency band available. If so, allocate an idle frequency band to it, and then proceed to the next step; if not or A itp2 >η2, then go to step 4.

8. Step 4.6: Put it into the execution queue Sche(j) and Delete from the task request queue, i=i+1, The residence time is stored in t_store, η2 = η2-A itp2 , otherwise go directly to the next step; Step 4.7: itp2 = itp2 + 1. If itp2 ≤ X2 and η2 > 0, return to step 4.4, otherwise proceed to the next step. Step 4.8: The scheduling analysis at time tp2 is completed, and the time pointer tp2 = tp2 + max(t_store) is updated, and η2 = A total ; Step 4.9: If tp2 ≥ t end Or i = N2, then the analysis on the right side of this scheduling interval ends; Step 5: Set s=s j The scheduling sequence Sche(j) at that time is saved, and the cost Cos(j) of selecting Sche(j) as the final scheduling sequence is calculated using formula (6), where the cost Cos(i,j) of task i is calculated using formula (7), j = j + 1; C dp is the penalty factor for task loss, and d describes the impact of priority on the cost of task loss; Step 6: If j>L, then set s for the current scheduling interval The corresponding scheduling sequence is the optimal scheduling sequence, and the scheduling interval analysis ends; otherwise, return to step 2.

Images