Journal of Computatonal Informaton Sytem 7:2 (211) 385-393 Avalable at http://www.jofc.com Multfuncton Phaed Array Radar Reource Management: Real-me Schedulng Algorm Janbn LU 1,, Hu XIAO 2, Zemn XI 1, Mngmn ZHANG 1 1 Electronc Engneerng College, Naval Unverty of Engneerng, Wuhan 4333, Chna 2 Ar and Space Borne Radar Reearch Secton, Ar Force Radar Academy, Wuhan 4312, Chna Abtract In paper a real-tme ta model of multfuncton phaed array radar bult, and a novel chedulng algorm propoed. h algorm tae e prorty of ta functon mode and deadlne nto account ynetcally, and can adapt well to dfferent load condton. Baed on e ta model, e algorm can acheve e varety of cheduler tme load n a real-tme fahon and adjut e ta parameter correctly when e ytem over loadng. he mulaton reult how at e propoed algorm mprove e cheduler performance w decreang e med deadlne rate effectvely, and e adjutment trategy ratonal and effectve on ytem overload. Keyword: Phaed Array Radar; a Schedulng; Deadlne; Prorty; me Load 1. Introducton A a mult-functon and hgh-performance radar ytem, phaed array radar have e advantage of flexble beam pontng drecton, veratle waveform, controllable ytem parameter, a well a e effectve reource allocaton trategy and powerful data proceng capacty. It playng an mportant role n e future advanced radar ytem[1]. Whle all e aforementoned predomnance depend on t effectve reource management. Reource management technque for phaed array radar am at mprovng radar ytem performance by effectvely ta chedulng and parameter control. here are many chedulng tratege for phaed array radar uch a fx-template, mult-template, partal template and adaptve chedulng, where e adaptve chedulng e mot effectve and complex n e real applcaton[2]. Many recent tude[3-6] have dealt w reource management problem for phaed array radar ytem ung real-tme dwell chedulng technology, but two lmtaton ext. he frt one e correlaton eparaton of dfferent radar dwell from e ame ta. For a new ta, t not clear at wheer e cheduler ha enough reource to allocate. When e ytem overloaded, e deleton of ome ta wll occur. he econd lmtaton at only e ta mportance condered, regardle of t urgency attrbute. h wll degrade e cheduler performance. he concept of radar dwell tme wndow preented n [7] whch denote e effectve pan of dwell requet. Kuo [8,9] propoe a rate-baed approach to chedulng radar dwell n a real-tme fahon. It reerve radar reource for all ta Correpondng auor. Emal addree: lu_jan_bn@163.com (Janbn LU). 1553-915/ Copyrght 211 Bnary Informaton Pre February, 211
386 J. Lu et al. /Journal of Computatonal Informaton Sytem 7:2 (211) 385-393 neceary for e mnmal radar operaton. But e real dwell tme wndow neglected, and e trategy to deal w ytem overloaded n t dcued. In paper, a novel real-tme ta model bult for phaed array radar, n whch each nd of dwell requet are combned a a radar ta. h ta model can be effectvely appled n e radar cheduler degn. Furermore, a real-tme chedulng algorm preented baed on e propoed ta model. 2. Adaptve a Model he propoed adaptve radar ta model nclude e parameter of arrvng tme, deadlne, tranmon tme, dwell leng and functon prorty of every dwell requet. 2.1. Search a Model Baed on e pror nformaton, phaed array radar can dvde e whole urvellance pace nto ome maller regon w e optmal earch eory. hen dfferent parameter can be adopted for dfferent earch regon o a to maxmze e radar earch performance. Aumng ere are N regon n whole urvellance pace, e (=1,2,,N) regon nclude B n beam poton w dwell leng Δ t n each beam poton. he earch frame tme and functon prorty of e regon are P and pr, repectvely. hen e ta model of earch regon can be expreed a, j, j aj dj ej = { = 1, 2,, }, { t, t, t, t, pr } n j B = Δ, ( = 1, 2,..., N ) (1) where, j denote e j dwell requet of e earch regon, t aj, dj t and t ej are t correpondng arrvng tme, deadlne and tranmon tme. Wout lo of generalty, for all e requet n e ame earch regon equal dwell leng and functon prorty are adopted. Before phaed array radar tart to earch ome gven regon, all correpondng dwell requet can be determned baed on e reult of earch parameter optmzaton and beam poton arrangement. o maxmze e flexblty of earch ta, we can aume at all requet arrve n turn at e begnnng of each earch frame, at a1 t = t, ( = 1, 2,..., N ) (2) aj a( j 1) t = t +Δ t,( j = 2,3,, Bn) Smlarly, e deadlne of each requet atfe dj t = t + P ( Bn j) Δ t, ( = 1, 2,..., N; j = 1, 2,, Bn) (3) where t denote ome reference tme, at tme all earch ta begn to execute (equvalently radar boot-trap). Equ. (2) and (3) how at n each earch frame all dwell requet need be accomplhed n tme order. When one frame end, e mlar earch dwell requet are generated at e begnnng of next earch frame. W e earch ta model propoed above, e frequency of earch ta
J. Lu et al. /Journal of Computatonal Informaton Sytem 7:2 (211) 385-393 387 n f = B / P (4) When e earch regon fxed, accordng to equ.(4) e frequency of earch ta determned by e earch frame tme. he bgger n amount of P, e leer of earch dwell requet, e leer of e load of e radar ytem and vce vera. For e real phaed array radar, e earch frame tme correlatve w e radar functon, and e mnmum of whch e um of e dwell leng for e total requet.,mn n P = B Δ t (5) he electon prncple for e maxmum P at e radar ytem can detect e target whch overpa e earch regon w e gven detecton probablty. For e earch regon w e elevaton θ, one target overpae regon w e velocty v at e range R (a fg.1). hen e tme pent for e target to overpa e regon δt = R θ /( v n α ), and e radar hould rradate e target twce at leat for e enough detecton probablty. hat expreed a δ t 2P and e maxmum P can be P,max = R θ /(2v n α) (6) v α θ R Fg.1 he Parameter of e Search Regon and e arget In e real radar chedulng proceng, e earch frame tme P [ P,mn, P,max ] can be adjuted flexbly to change e load of e correpondng earch ta, whch can mae e earch ta adapt to e radar ytem reource. 2.2. rac a Model For e multfuncton phaed array radar, each trac ta correpond to a target, and e type of trac ta nclude normal trac, prece trac, hgh prece trac, mle gudance, etc. When upcou target are detected n e earch mode, a confrmaton ta generated n e drecton of e target to verfy t preence. Once a target dentfed, e correpondng confrmaton ta no longer needed. Intead, a equence of perodc-le trac dwell generated to trac e target. Suppoe at ere are K type of trac ta n radar ytem, where e type of trac ta nclude M target (=1,2,,K), e trac ta model of e target ( = 1, 2,..., M )
388 J. Lu et al. /Journal of Computatonal Informaton Sytem 7:2 (211) 385-393 where n,, j aj dj ej = { = 1, 2,, }, { t, t, t, t, pr } j n j = Δ, ( = 1, 2,..., K; = 1, 2,..., M ) (7) denote e number of trac dwell requet for e target (equvalently target tracng pan),but t unnown a e capture tme and dappearng tme of e target can t be determned. he oer parameter are mlar to e earch ta model. Δ denote e tracng ample nterval for e target, and wout lo of generalty, aumng one confrmaton dwell before target tracng, en n = ( td tcap )/ Δ (8) a1 t = tcap aj a( j 1), ( 1, 2,..., K; 1, 2,..., M) t t = = = +Δ (9) t t t t t t d1 a1 = +Δ cw dj aj = K = M = +Δ w, ( 1, 2,..., ; 1, 2,..., ) (1) where x denote e maxmal nteger whch le an x, t cap and t d are capture tme and dappearng tme for target, Δ tcw and Δ t w denote e tme wndow of confrmaton ta and e type trac ta. W e trac ta model above, e frequency of trac ta f = 1/ Δ (11) where f denote e tracng ample frequency and equal to e frequency of e trac dwell requet for e target. he tracng ample nterval Δ can be elected wn e ratonal range, e mnmum of whch determned by e hardware and oftware of e radar ytem, for example Δ,mn SI where SI denote e chedulng nterval for phaed array radar. And t maxmum correlated w e character of tracng ta. For example, e maxmum can be 2 more or le for e trac mantenance of normal target, whle for e target w hgh reaten prorty e maxmum mut be retrcted to a horter tme range a.5. 2.3. Oer a Model h type of ta manly nclude trac lo ta, elf-examnaton ta, calbraton ta, pecal experment ta, etc. Uually, oe ta only conume one or everal contnuou beam dwell tme, whch are le an earch ta or trac ta n quantty and undetermned for e radar cheduler. So an ndvdual model hould be etablhed for each ta. Aumng H type of oe ta, e ta model,1 a1 d1 e1 R R = { R }, R { tr, tr, tr, t, pr },1 R = Δ, ( = 1,2,..., H) (12) he parameter of e ta model n equ.(12) are mlar to oe of e earch or trac ta. he dfference at ta ha only one requet. In equ.(12), a1 R d R t and t 1 are unnown, whle R Δ t and R pr vary w e dfferent ta.
J. Lu et al. /Journal of Computatonal Informaton Sytem 7:2 (211) 385-393 389 Above ree ta model cover mot phaed array radar functon. Whle e rd type of ta tochatc and conume only a lttle of ytem reource, e frt two type of ta are e man factor affectng e cheduler effcency. 3. Adaptve Schedulng Algorm Accordng to above radar ta model, we now at e mot ta n phaed array radar ytem are aperodc and non-preempt (one ta durng executed coure can t be nterrupted by oer). In a word phaed array radar ta chedulng belong to non-preempt hard real-tme chedulng problem. h problem NP-Hard, and t oluton may be not exted or unque. So e heurtc meod uually ued to obtan t uboptmal oluton[1-11]. 3.1. Degn Meod of a Prorty In paper two charactertc parameter of ta, relatve deadlne and functon prorty, are ued to degn t ntegrated prorty. he bac prncple of ta chedulng are: (1) e hgher of functon prorty of dwell requet, e hgher of e fnal prorty ; (2) e earler of relatve deadlne, e hgher of e fnal prorty. Aumng at ere are totally Q dwell requet n e cheduler currently, denoted a q = { q1, q2,..., q Q } whch atfy: (1) e arrvng tme of all dwell requet no more an e current tme; (2) e relatve deadlne of each requet more an t dwell leng. Sort e above dwell requet by functon prorty from hgh to low and by relatve deadlne from early to late, and we can get two requet chan, functon prorty chan and relatve deadlne chan, repectvely. he equence number of each requet n e two chan can be obtaned, whch are functon prorty equence number Np and relatve deadlne equence number Nd. Obvouly ee two equence number atfy Np, Nd [ 1, Q]. hen, e ntegrated prorty of each ta can be obtaned rough p = f ( Np, Nd) (13) Above functon f can be elected accordng to ome pecal ntenton, whle e mplet form e lnear functon adopted here. p = [ η Np + ( Q+ 2 η) Nd]/( Q+ 1) (14) where e factor η le between 1 and Q + 1. From equ.(14) ome tradeoff made between functon prorty and relatve deadlne. And for e pecal cae η = Q /2+ 1, equal mpact of ee two factor on e fnal prorty, we call chedulng trategy a HPEDF. One pont to menton here at, e fnal prorty calculated w equ.(14) may not correpond to each requet one for one. When multple requet have e ame fnal prorty, FIFO (Frt Come Frt Out) rule can be ntroduced to chedule em. 3.2. Realzaton of Adaptve Schedulng Algorm Dfferent from e general real-tme ytem, e ta chedulng n phaed array radar ytem executed accordng to ome fxed tme nterval, whch called chedulng nterval. he dwell requet n e next chedulng nterval are analyzed n e current chedulng nterval. Suppoe totally L dwell requet
39 J. Lu et al. /Journal of Computatonal Informaton Sytem 7:2 (211) 385-393 q = { q1, q2,..., q L } n e next chedulng nterval arrve at radar cheduler orderly, e realzaton of e current analy n radar cheduler a follow: Step 1. Obtan e tart tme ndex tp( tp t a ) and et = 1; Step 2. Delete oe dwell requet whoe relatve deadlne are le an e dwell leng. Aume e number of oe requet n, and et = + n ; Step 3. Fnd out all e requet whoe arrvng tme le an e current tme ndex tp, denoted a q = { q, q,..., q }. hen calculate e fnal prorte of ee requet; 1, 2, Q, Step 4. Select e requet q j, w e maxmal fnal prorty from q, and chedule dwell requet uccefully; Step 5. et tp equal to e fnhng tme of requet j q, and et = + 1. If tp > ta + SI or > L, go to tep 6, ele return to tep 2; Step 6. chedulng analy end, get e whole cheduled dwell requet and tme ndex tp. In e chedulng procedure above, e lat dwell cheduled maybe occupy ome pan of next chedulng nterval, becaue of e non-preempt charactertc of dwell requet. So e tme ndex tp need to be reerved for e analy n e next chedulng nterval. 4. Scheduler Load Analy 4.1. me Load of Scheduler In ecton, only earch and trac ta are analyzed for cheduler worload, a e rd nd of ta o few at t worload can be omtted. Accordng to e ta model n ecton 2, we can get e rato of e earch ta occupyng radar ytem tme ζ = B Δ t / P,( = 1,2,..., N) (15) n where e parameter ζ denote e earch ta conume ytem reource rato averagely. Becaue e phaed array radar can trac multple target and earch e gven regon at e ame tme w e AS(rac And Search) mode, e occupy rato of e target trac ta can be expreed a ζ =Δt / Δ, ( = 1, 2,..., K; = 1, 2,..., M ) (16) hu, e total tme rato of all earch and trac ta, called a e cheduler tme load N K M = 1 = 1 = 1 (17) ζ = ζ + ζ Inequalty ζ 1 denote e cheduler load ratonal, but only a neceary condton for chedulng all requet. Even f ζ 1, ome requet tll may be deleted for e conflcton n tme. But ζ > 1 denote cheduler overloaded, whch ndcate ome requet deleted defntely n e proce of ta chedule.
J. Lu et al. /Journal of Computatonal Informaton Sytem 7:2 (211) 385-393 391 4.2. Adjutment Meod for Scheduler Overload Aumng current cheduler tme load ζ < 1, a new ta (earch or trac), whch load ζ, add n e cheduler. If ( ζ + ζ ) 1, ta can be permtted to be cheduled; oerwe ta wll reult n overload. In ntance, we need ome adjutment trategy to tradeoff between ta and oer, uch a reducng e tme load of ta or oer. Aumng e tme load of e appendng ta after adjutment ' ζ, oer target trac ample rato after adjutment Δ ', each earch frame tme ' P, oe parameter mut atfy N K M ' ' ' Bn t P t ζ = 1 = 1 = 1 Δ / + Δ / Δ + 1 (18) Equ.(18) ndcate e cheduler load after adjutment tll need be le an 1. W above trategy, e tme load of radar ytem can be ept n a ratonal cope. Note at, bede earch frame and trac ample rato, e oer ta parameter can alo be adjuted w meod. 5. Smulaton and Analy 5.1. Smulaton Parameter Accordng to e radar ta model propoed, we elect e man four wor mode of phaed array radar, whch are earch, trac, trac confrmaton and trac lo ta. Furermore ere are two type for earch ta, whch are volume earch and horzon earch. For e trac ta, ere are alo two type adopted, normal trac and prece trac. And e rato of target number w above two type of trac ta 4:1. he oer parameter are hown n table I, where e perod denote e earch frame tme for e earch ta or e tracng ample nterval for e tracng ta. Smulaton tme 12 econd, and e chedulng nterval et to be 5 m. he trac confrmaton ta are whle a new target detected, and e trac lo ta are produced randomly w a gven probablty 1%. In our mulaton, capture tme of each target choen randomly between tart tme of mulaton and e frt trac ample perod, and e dappearng tme aumed to be e end of mulaton. he reult below are obtaned rough 1 tme mulaton. 5.2. Smulaton Reult and Analy We elect e med deadlne rato (MDR) a e man evaluaton parameter for e chedulng algorm. he MDR defned a e rato between e number of all deleted dwell requet and e total number of all requet. It nverely proportonal to e chedulng ucce rato (SSR). he lower e MDR, e hgher e SSR. he chedulng algorm propoed n paper compared w e tradtonal meod n whch all dwell requet are cheduled by e functon prorty, at e dwell requet w e hghet functon prorty proceed frt. Frtly, e fxed ta parameter are elected for e chedulng proceng, where e frame tme of volume earch and horzon earch are 4 and 2 repectvely, e ample nterval of normal trac and
392 J. Lu et al. /Journal of Computatonal Informaton Sytem 7:2 (211) 385-393 prece trac are 1 and.5. hen fg.2(a) how e tme load of cheduler vare w e target number traced. When e target number equal to 5, e worload of radar cheduler aturated. ID a ype able 1 Phaed Array Radar a Parameter Functon Prorty Dwell Leng me Wndow Beam Number Perod 1 rac Confrmaton 6 4 m 3 m 1-2 Prece rac 5 2 m 3 m 1.5~1 3 Normal rac 4 4 m 5 m 1 1~2 4 rac Lo 3 4 m 5 m 4-5 Horzon Search 2 8 m - 1 2~2.5 6 Volume Search 1 4 m - 4 4~5 Furermore, fg.2(b)(c) gve e chedulng reult wout adjutment when e ytem overloaded. It hown at e tradtonal meod can eep e hgh functon prorty ta beng cheduled effectvely. But large number of low functon prorty ta are deleted. he propoed HPEDF algorm can eep almot all trac ta cheduled, whle deletng fewer earch ta. 1.3.2.1 tme load of cheduler 1.2 1.1 1.9.8 med deadlne rato of earch ta.18.16.14.12.1.8.6.4 e propoed meod e tradtonal meod med deadlne rato of trac ta.9.8.7.6.5.4.3.2 e propoed meod e tradtonal meod.2.1.7 (a) me Load Veru arget Number (b) MDR of Search a (c) MDR of rac a Fg. 2 Schedulng Relut wout Parameter Adjutment Secondly, e parameter of e earch and trac ta can be changed adaptvely from table I when e target number more an 5, howed n fg.3(a). Accordng to equ.(18), e earch frame tme of volume earch adjuted frtly when e radar ytem overloaded. A e earch frame tme ncreaed to 5, e cheduler can wor normally w e no more an 7 target. Furermore, e frame tme of horzon earch can be adjuted when e target number between 7 and 9. Lat, e ample nterval of normal trac and prece trac can be ncreaed for e more target number, and e tme load of e ytem doen t exceed t upper lmt 1 all e tme. Accordng to aforementoned parameter adjutment cheme, chedulng reult are hown n fg.3(b)(c). Comparng fg.3 w fg.2, an mprovement can be een obvouly. he propoed algorm can chedule all requet, whle e tradtonal meod tll need to delete a few earch ta. From above mulaton reult and analy, ome concluon can be drawn: (1) e adjutment cheme can effectvely mprove e performance of radar cheduler; (2) e propoed algorm can ynetcally conder e functon prorty and e relatve deadlne, o t performance better an e tradtonal
J. Lu et al. /Journal of Computatonal Informaton Sytem 7:2 (211) 385-393 393 meod. 6.2.1 ample nterval frame tme frame tme 5 4 Volume Search 3 3 2 1.5 Horzon Search 1 1 Normal rac.5 med deadlne rato of earch ta.18.16.14.12.1.8.6.4.2 e propoed meod e tradtonal meod med deadlne rato of trac ta.9.8.7.6.5.4.3.2.1 e propoed meod e tradtonal meod (a)parameter Adjutment (b) MDR of Search a (c) MDR of rac a Fg.3 Schedulng Reult w Parameter Adjutment 6. Summary and Concluon Many extng phaed array radar ytem tll adopt neffcent or even non-real-tme reource management technque, uch a FIFO-le or cyclc-executve-le chedulng algorm. A a reult, much radar reource wated wout gnfcant performance mprovement. Baed on real-tme eory, paper am at e eental ue for e degn of modern phaed array radar reource management - ta chedulng. A novel real-tme ta model bult and correpondng chedulng algorm propoed for phaed array radar ytem. Smulaton reult how at e propoed algorm can effectvely chedule radar ta n a real-tme fahon w optmal performance. Reference [1] D.R. Blletter. Multfuncton Array Radar. Norwood, MA: Artech Houe, 1989. [2] M.L. Baugh. Computer Control of Modern Radar. RCAM&SRMooretown Lbrary, 1973. [3] Dan Stromberg, Per Grahn. Schedulng of ta n phaed array radar[c]. Proc. IEEE Internatonal Sympoum on Phaed Array Sytem and echnology, 1996, pp:318-321. [4] G.-G. Lee, P.-S. Kang, C.-S. Shh, L. Sha. Radar dwell chedulng conderng phycal charactertc of phaed array antenna[c]. Proc. IEEE 24 Real-me Sytem Sympoum, 23, pp:14-24. [5] S. Gopalarhnan, C-S. Shh, P. Gant. Radar dwell chedulng w temporal dtance and energy contrant[c]. Proc. Internatonal Conference on Radar Sytem, 24, pp:1-4. [6] S.L.C. Mranda, C.J. Baer, K. Woodbrdge. Phaed array reource management: a comparon of chedulng algorm[c]. Proc. IEEE Radar Confernece, 24, pp:79-84. [7] Huzng, A.G., Bloemen, A.F. An effcent chedulng algorm for a multfuncton radar[c]. Proc. IEEE Radar Conference, 1996, pp:359-364. [8] e-we Kuo, Yung-Sheng Chao, Chn-Fu Kuo. Real-me Dwell Schedulng of Component - Orented Phaed Array Radar[J]. IEEE ranacton on Computer, 25, 54(1): 47-6. [9] e-we Kuo, Yung-Sheng Chao, Chn-Fu Kuo. Real-me Dwell Schedulng of Component - Orented Phaed Array Radar[C]. IEEE Radar Conference, US:IEEE, 22, pp:92-97. [1] Hatao Zhang, Songcan Zhang. Schedulng analy of dtrbuted real-tme embeded ytem[j]. Journal of Computatonal Informaton Sytem, 21, 6(7): 2373-2382. [11] K. Jeffay, D.F. Stanat, C.U. Martel. On non-preemptve chedulng of perodc and poradc ta[c]. Proc. 12 IEEE Real-me Sytem Sympoum, 1991, pp:129-139.