Design of a Front End Amplifier for the Maximum Power Delivery and Required Noise by HBMO with Support Vector Microstrip Model

Transcription

1 34 F. GÜNEŞ,. DEMİRE, P. MAHOUTİ, DEIGN OF A FRONT END AMPIFIER FOR THE MAXIMUM POWER DEIVERY Desgn of a Front End Amplfer for the Maxmum Power Delvery and Requred Nose by HBMO wth upport Vector Mcrostrp Model Flz GÜNEŞ, alh DEMİRE, Peyman MAHOUTİ Department of Electroncs and Communcaton Engneerng, Yıldız Techncal Unversty, Davutpasa Campus, 34, Istanbul, Turkey gunes@yldz.edu.tr, salhd@yldz.edu.tr, pmahout@gmal.com Abstract. Honey Bee Matng Optmzaton (HBMO) s a recent swarm-based optmzaton algorthm to solve hghly nonlnear problems, whose based approach combnes the powers of smulated annealng, genetc algorthms, and an effectve local search heurstc to search for the best possble soluton to the problem under nvestgaton wthn a reasonable computng tme. In ths work, the HBMO-based desgn s carred out for a front-end amplfer subject to be a subunt of a radar system n conjuncton wth a cost effectve 3-D ONNET-based upport Vector Regresson Machne (VRM) mcrostrp model. All the matchng mcrostrp wdths, lengths are obtaned on a chosen substrate to satsfy the maxmum power delvery and the requred nose over the requred bandwdth of a selected transstor. The proposed HBMO-based desgn s appled to the desgn of a typcal ultra-wde-band low nose amplfer wth NE35 on a substrate of Rogers 435 for the maxmum output power and the nose fgure F(f) = db wthn the 5- GHz usng the T-type of mcrostrp matchng crcuts. Furthermore, the effectveness and effcency of the proposed HBMO based desgn are manfested by comparng t wth the Genetc Algorthm (GA), Partcle warm Optmzaton (PO) and the smple HBMO based desgns. Keywords Honey Bee Matng, ow Nose Amplfer, mcrostrp, optmzaton, matchng crcut, VRM.. Introducton Consderng all the strngent requrements whch nclude hgh gan, low nput and output Voltage tandng Wave Rato (VWR) s, low nose fgure together wth the low-power consumpton from the low-level battery, the wdeband mnature ow Nose Amplfer (NA) desgn s one of the bggest challenges to Ultra -Wdeband (UWB) transcever ntegratons. To meet these strngent requrements, frst of all, fast and low-nose, hgh-qualty transstors are needed. In fact, today s semconductor technology has been focusng on producng the mcrowave transstors wth the ntrnsc superor frequency characterstcs. econd ssue s of course to establsh the compromse nterrelatons among the power gan, the nput/output VWRs, the nose fgure, the bas condtons (V D, I D ) and frequency (f) of the two port transstor. Recently, the nonlnear performance equatons of the transstor are solved smultaneously wth respect to the source mpedance n the [z]-doman for the maxmum power delvery and the requred nose usng the lnear crcut and nose theores, by our research group to be used n the desgn of the front-end amplfer [], []. Thus dependences of the maxmum gan G Tmax under the conjugate matched output s obtaned on the rgorous mathematcal bases wth respect to the nose fgure, nput VWR throughout operaton doman (V D, I D, f). On the other hand, one of the recently proposed nature nspred ntellgence algorthms that have shown great potental and good perspectve for the solutons of varous dffcult optmzaton problems s the HBMO [3-6]. The HBMO algorthm frst proposed by Afshar et al. has been used to solve a sngle reservor optmzaton problem [3], [4], clusterng [5], state estmaton n dstrbuton networks [6]. In ths work, we propose a HBMO desgn optmzaton procedure gven n Fg. for a front-end amplfer so that all the matchng mcrostrp wdths, lengths { W, } can be obtaned to provde the (Z, Z ) termnatons on a gven substrate (ε r, h, tanδ) for the maxmum power delvery and the requred nose over the requred bandwdth of a selected transstor, respectvely. The HBMO procedure of the front-end amplfer desgn (Fg. ) can be consdered to be consstng of the followng stages: () Frstly, Feasble Desgn Target (FDT) s bult by solvng analytcally [], [] or numercally the hghly nonlnear performance equatons of the transstor for the maxmum power delvery and the requred nose to determne the necessary source Z and load Z mpedances of the actve devce at a chosen bas condton (V D, I D ) of the devce as follows:

2 RADIOENGINEERING, VO. 3, NO., APRI 4 35 Z ( ) r ( ) jx ( ), () Z ( ) ( ) ( ) * r jx Z out ( Z ( )), () where r ( ), r ( ),,..., N (3) and the proposed HBMO desgn s resulted wth ts outstandng performance, besdes the verfcaton s also made usng the crcut smulator AWR. TART Transstor: [], [N] Operaton Bandwdth: f lower,f upper F req(f) > F mn(f) ubstrate: ε r, h, tanδ Mcrostrps mts: w mn, w max, l mn, l max Performance charactersaton Programme {Z (ω ), Z (ω )=Z * out(ω )} for the {F req(ω ), V out(ω )=, G Tmax(ω )} trplet VRM Model of Mcrostrp Z ε o Eff Calculate Cost t max=n GEN, N Drone, Cost req Cost < Cost req t < t max Yes Fnal oluton s W, =,, 3 No Updatng Process by HBMO END Fg.. Flow chart of the HBMO+VRM desgn optmzaton procedure. () econd stage s the desgn optmzaton of the mcrostrp wdths, lengths { W, } of the nput/output matchng crcuts usng the HBMO wth the royal jelly, to provde the necessary source Z (ω ) and load Z (ω ) mpedances, respectvely to the devce. Thus, n ths desgn optmzaton procedure, the mcrostrp wdths and lengths { W, } on a selected substrate (ε r, h, tanδ) are drectly used by the HBMO algorthm (Fg. ) and the cost functon s evaluated by means of the VRM mcrostrp model (Fg. 3). The 3-D ONNET based VRM model of the mcrostrp [7], [8] s employed that provdes an accurate, fast and cost effectve generalzaton from the hghly nonlnear dscrete mappng from the nput doman M(R 4 ) of the mcrostrp wdth, substrate (ε r, h) and frequency (f) to the output doman of ether the characterstc mpedance Z or effectve delectrc constant ε eff. () Fnally the proposed HBMO based desgn s appled to the desgn of a typcal ultra-wde-band NA wth NE35 on a substrate of Rogers 435 (ε r = 3.48, h =.54 mm, tanδ =.3) for the maxmum output power and the nose fgure F(f) = db wthn the 5- GHz usng the T-type of mcrostrp matchng crcuts. Furthermore the completed amplfer desgn s also compared wth the GA, PO and the smple HBMO desgns Fg.. Flow chart of the HBMO algorthm. Fg. 3. The cost effectve VRM modelng of mcrostrp lnes. The artcle s organzed as follows: ectons and 3 gve the objectve and varables of the desgn optmzaton of the matchng crcuts. The HBMO wth the royal jelly feed algorthm takes place n ecton 4. As a test vehcle of the presented methodology, desgn of typcal wdeband low-nose amplfers usng T-type of mcrostrp matchng crcuts s dscussed comparatvely wth the GA, PO and the smple HBMO desgns n ecton 5 wth together ther

3 36 F. GÜNEŞ,. DEMİRE, P. MAHOUTİ, DEIGN OF A FRONT END AMPIFIER FOR THE MAXIMUM POWER DEIVERY valdated performances. Fnally the paper ends wth conclusons n ecton 6.. Desgn Objectves. Objectve for Determnaton of FDT In ths desgn optmzaton problem, the desgn objectve s the maxmum output power delvery and the requred nose. For ths purpose, frstly the gan G T s maxmzed wth respect to the Z under the matched output port provded that requred nose F s satsfed, whch can be expressed as: load Z mpedances of the actve devce at a chosen bas condton (V D, I D ) of the devce as follows: P G T G AV (Z ) M out (Z, Z ) P (4) avs where z (Z ) r G AV r z Z out Z * (Z, Z ) out Z M out Zout Z z z Z z out z Z (5) R n Z Z N opt F( Z ) F mn (6) N o Zopt r Here z j,, j =, and R n, F mn and Z opt = r opt + jx opt are respectvely, the z-parameters obtaned by converson of the []- parameters and nose [N]- parameters at an operaton condton {(V D, I D ), f}. Thus, hereafter the problem of determnaton of the source mpedance Z = r + jx of a mcrowave transstor can be descrbed as a mathematcally constraned optmzaton problem so that the gan G AV (r, x ) gven by (5) wll be maxmzed and smultaneously the requred nose fgure wll be met usng the F(r,x ) equaton (6) at each sample frequency throughout the requred operaton bandwdth. Thereby the mult-objectve cost functon of ths constraned optmzaton process can be expressed as: Cost F x f F f (7) G AV (r, x e, f ) (r,, ) req( ) wth the followng constrants e Z, e Z. (8) z z e Z e z n out { } z Z { z zz } z Z e Z e Freq mn (9) F. () In (7), Ψ and Ψ are the weghtng coeffcents whch can be chosen durng the optmzaton process by tral, whch n our case are taken as unty. Thus, the smaller cost s the ftter optmzaton process we have.. Desgn Objectve of the Matchng Networks Thus, we have the transstor termnatons gven by () and (). In the desgn optmzaton procedure, the gan of the nput/output matchng two-port termnated by the complex conjugate of the obtaned n the prevous subsecton, s maxmzed over the requred bandwdth: cost ( ) Mnmum (- G T ( f, )) () where s the desgn varable vector whch conssts of the mcrostrp wdths and lengths of the problem matchng crcut and G T s the power gan of the same matchng crcut at the sample frequency f. In the worked example T-type matchng crcuts are consdered to be desgned. The proposed method can be appled wthout any dffculty to another dfferent type of matchng crcut such as - or -types or any other knd of matchng crcuts. In that case, the gan functon G ( f, T ) gven n () should be evaluated for the consdered matchng crcut. 3. Desgn Varables: Mcrostrp Wdths and engthsw, Desgn varables are the mcrostrp wdths, lengths { W, } of the nput/output matchng networks on a selected substrate {ε r, h, tanδ}, whch are mapped n the contnuous manner to the characterstc mpedance Z and the delectrc constant ε eff of the equvalent transmsson lne to be used n the desgn optmzaton process va the two 3-D ONNET-based VRMs [7], [8]. Here, the nput doman of the mcrostrp VRM model s four-dmensoned M(R 4 ) wthn {. mm W 4.6 mm, ε r,. mm h. mm, GHz f 4 GHz} and the output domans of the Z and ε eff correspond to {3 Ω Z 4 Ω} and {.5 ε eff 9.7}, respectvely. The mathematcal bases of upport Vector Regresson can be found n [7] n detals.

4 RADIOENGINEERING, VO. 3, NO., APRI Buldng Knowledge-Based Mcrostrp VRM Model Knowledge-based Mcrostrp VRM s gven as block dagram n Fg. 3 where the quas-tem mcrostrp analyss formulae [] s used as a coarse VRM model data base by means of whch nfreq n nhnw554 number of ( x, y ) data pars are obtaned to tran the coarse VRM, where n freq, n ε, n h, n w are the number of the samples for the frequency, the delectrc constant, the substrate heght and wdth, respectvely. Tab. gves the accuracy of the Z coarse model wth the number of the Vs and the radus of selecton tube ϵ. 4 and 367 fne Vs obtaned from 3-D ONNET smulator are used to tran the fne Z and ε eff VRMs, respectvely wth the accuracy at least 99.4 % (Fg. 4b). Thus the substantal reducton (up to 6 %) s obtaned utlzng sparseness of the standard VRM n number of the expensve fne dscrete tranng data wth the neglgble loss n the predctve accuracy and the resulted fne mcrostrp VRM model can be consdered as accurate as the 3-D EM smulator [9] and as fast as the analytcal formulae []. The typcal comparatve predcton curves of the mcrostrp VRM model take place n Fgs. 4a and 4b whch gve varatons of the characterstc mpedance Z and the effectve delectrc constant ε eff wth mcrostrp wdth W resulted from the fne VRM model for varous delectrcs wth h =.8 mm at 4 GHz, respectvely. Epslon (ϵ) Number of Vs Accuracy (%) Tab.. Accuracy of the coarse VRM model. In the next secton, HBMO wth Royal Jelly algorthm wll be gven to determne the matchng mcrostrp wdths, lengths { W, } on a chosen substrate {ε r, h, tanδ} to satsfy the requred nose and the maxmum power delvery over the requred bandwdth of a selected transstor. The FDT s determned usng the smple HBMO verson wthout the Royal Jelly. 4. HBMO wth Royal Jelly for the Amplfer s Matchng Network Desgn Problem 4. Workng tages of the Proposed HBMO The HBMO wth Royal Jelly for the desgn of the T- type mcrostrp front-end amplfer problem can be descrbed n the followng stages (Fg. ): tage : Defnton of nput data In ths stage, the number of the Drone bees (N Drone ), maxmum teraton number (t max ), szes of the genetc n- Z () eff Analytcal Formulatons 3D onnet mulaton Fne Predct Wdth (mm) (a) Analytcal Formulatons 3D onnet mulaton Fne Predct Wdth (mm) (b) Fg. 4. Comparatve varatons of (a) Characterstc mpedance Z, (b) Effectve delectrc constant ε eff vs wdth of the analytcal formulatons, Fne Model and the 3-D ONNET smulaton on the substrate ε r = 3.48, h =.54 mm at f = 8 GHz. hertance of the Master Queen Q M and each Drone bee D j, (m Q, m D ); maxmum number of feedng tmes of the Master Queen Q M wth Royal Jelly (N RJ ), Maxmum (E max ) and mnmum (E mn ) energes of the Queen at the start and end of the matng flghts, respectvely, and the requred cost cost req are defned by the user. In the algorthm, the numbers of the Hve (N Hve ), Brood (N Brood ), arva (N arva ), Fertlzaton (N fertlzaton ) are set equal to (N Gen )whch s taken to be equal to t max and the total egg number N Egg = (N Gen ) 5. tage : Defne Queen Q and Drone s D populatons wth ther genetc nhertances Q _ populaton [ Q,,..., ] t Q Q N Hve () D _ populaton [ D, D,..., D N ] t Drone where Q and D j members of the Queen and Drone populaton are defned based on the optmzaton varables, respectvely as below: wq, wq, w3 Q Q Q, Q, 3 Q x3 (3) wdj, wdj, w3dj D j Dj, Dj, 3Dj x3

5 38 F. GÜNEŞ,. DEMİRE, P. MAHOUTİ, DEIGN OF A FRONT END AMPIFIER FOR THE MAXIMUM POWER DEIVERY Furthermore genetc nhertance belongng to each Queen Q and Drone D j are also defned n terms of the optmzaton varables as follows: Q Gen W, W, W W, W, W 3 3 DjGen,, 3 3,, x 3 x3 (4) W = [w ] mx, = [ ] mx, =,,3 and m Q = for Queens and m D = for each Drone bee as the userdefned parameters. The genetc nhertance of a Queen bee Q Gen n (4) s completely passed to the next generaton whle only D jgen belongng to the N Drs drones whch had successful matng flghts passed down to the next generaton by beng collected nto the Queen Bee s spermatheca. In the next stage, ntalzaton of the Queen and Drones populaton and generaton of the Genetc nhertances wll be gven. tage 3: Intalze the Queen and Drones populaton and ther genetc nhertances Intalze all the elements of the Q_populaton to zero excludng the frst one: Q =,, the frst Queen wll be called hereafter as the Master Queen Q M that wll gve brth to the members of all the hves. The Master Queen Q M, the D_Populaton and ther genetc nhertances defned n (-4) are randomly ntalzed as follows: w [CC ]rand(.) Crand(.), (5) [C C ]rand(.) C rand(.). (6) where C, =, 3 and =, 4 are the gven upper and lower boundares for the wdths and lengths of the mcrostrp lnes, respectvely. Rand (.) s a random generator whch generates random values (, ). Then calculate Ftness Values of Master Queen Q M and the D_ Populaton n () usng ether (7) or (8-). In the next stage, spermatheca of the Master Queen Q M wll be generated. tage 4: Generate the Master Queen s spermatheca (Matng Flghts) At the start of the matng, Master Queen Q M fles wth her maxmum energy E max. A drone s randomly selected from the drone populaton n () and mates wth the Master Queen Q M probablstcally usng an annealng functon as follows: where f f f QM D j, f Et () Prob(Q M, D) e (7) f s Ftness of the Master Q M Queen Q M, f s Ftness of the j th Drone, D Dj j and the fnesses are evaluated usng the cost functon defned by (7) and (3). The calculated Prob(Q M, D) of the matng flght s compared wth a random value ϵ (, ) and f t s greater than the random value, the matng wll be assumed to be successful attempt and the correspondng Drone s sperms (Drone s genetc nhertances) wll be added to the spermatheca: Prob(Q M, D) Rand(.). (8) After each matng Master Queen s energy level s decreased wth the random decayng coeffcent: α(t) ϵ (, ), Et ( ) E( t)x ( t). (9) If the current energy level of the Master Queen Bee Q M s lower than the mnmum level, the matng flghts wll be stopped, thus the Master Queen Q M can fly back to the Hve to gve brths to new members of the colony as gven n the stages (5-6), otherwse the matng flghts wll be contnued untl there are no more drones n the D populaton to be mated. tage 5: Generate the Genetc Pool (GP) If a matng flght s successful, the Master Queen Bee Q M wll accept all the sperms of the partner drone nto her spermatheca to generate the genetc pool wth her genetc nhertance, from where new generaton of the entre colony wll have ther genetc denttes. ze of the genetc pool wll be ncreased wth the number of the successfully performed matng flghts, and the Master Queen Q M s genetc nhertance completely passes to next generaton whle solely the D jgen s belongng to the N Drs drones havng successful matng flghts tll the end of her matng flght, thus the genetc pool can be defned as W GP, W GP, W 3GP GP,,, GP GP GP x 3 W GP w,, 3 mx GP mx () Here m = + N Drs and N Drs N Drone s the number of the drones each of whch had a successful matng flght wth the Master Queen Bee Q M. Thus, n ths stage we have a search space havng a huge capacty wth (m) 6 = ( + N Drs ) 6 number of dfferent random solutons for the T-matchng crcut that means at least 8 solutons wth N Drs =. In the followng stage, generaton of Egg populaton (oluton pace) from ths genetc pool (earch pace) wll be gven by the crossng over process. tage 6: Generate the Egg populaton In the proposed HBMO algorthm, gender of all new born members of the colony wll be assumed as female, thus each soluton can be consdered as a potental Master Queen Bee canddate. The Egg_ populaton s defned as: Egg populaton [Egg, Egg,.., Egg ] t N Egg w Egg, w Egg, w 3Egg Egg Egg, Egg, 3Egg x3 ()

6 RADIOENGINEERING, VO. 3, NO., APRI 4 39 =,...,N egg, whch s generated by crossng over among the correspondng elements of the genetc pool: w jegg W (K n,) jgp (K ' GP n,) jegg j () K n and K n are randomly generated ntegers (, + N Drs ).Thus a sub-search space s generated as a soluton space by ths Egg populaton wthn the complete search space defned by the genetc pool wth the N egg = (N gen ) 5 =(t max ) 5 couples whch wll be sorted rapdly for the best soluton n the next stage. tage 7: Accelerated exploraton for the new Master Queen Bee Durng ths phase, all the new born members of the Egg Populaton wll be passng from the fve steps to explore rapdly the search space for the selecton of the new Master Queen Q M (The Best soluton). These fve steps can be gven wth ther szes as follows: -Fertlzaton (N fertlzaton ), -arva (N arva ), 3-Brood (N brood ), 4-Hve (N Hve ), and 5-Generaton (N Gen ), sze of each of these steps s equal to maxmum teraton number whch s taken to be equal to n our applcaton. The accelerated exploraton s based on the sortng step by step and can brefly be summarzed as follows: In each step, the current entre populaton s dvded nto the subpopulatons havng (N Gen ) members, then the best member wth the mnmum cost value of each sub-populaton s promoted to the next step, and the rest members are dscarded. For N Drs = 5, we have m = + x5 and 5 elements for each of W jgp and jgp vector of the genetc pool as gven n (), thus the complete search space conssts of 5 6 =.4 x 8 (W j, j ) couples, among whch N Egg = 5 = 3.. (W j, j ) pars are passed to the Egg populaton by crossng over to generate the soluton space. ortng these 5 sub-groups, total 4 eggs (the solutons havng the mnmum cost values belongng to the subgroups) are fertlzed and promoted to the arva stage. Then repeatng the same process n the arva stage, 3 larvas wll be promoted to the Brood stage. o on, at the end of the competton among the queens of the = 4 Hves, only best solutons wll be passed to the Generaton stage as the Master Queen Canddates. nce teraton s equal to Generaton number, Master Queen Canddate of each Generaton/Iteraton wll be compared wth the current Master Queen Bee. In ths fnal step, only (W j, j ) couples havng the mnmum cost of the competton wll be chosen as the new Master Queen Bee whch wll take new matng flghts to gve born to new members of the next generaton of the colony. These steps can be consdered as the accelerated selecton of the fttest member of the entre colony. The algorthm termnates f t reaches ether the maxmum teraton/generaton number or the requred cost. tage 8: Feedng of the Master Queen Bee wth Royal Jelly Royal Jelly Feed ams at ncreasng the accuracy of the soluton by mnmzng the cost obtaned from the global search. Therefore each element of the Master Queen Bee Q M s ether ncreased or decreased by the ncremental steps N RJ whle the other elements reman the same, dependng on the cost varaton. Thus, the accuracy of the soluton s ncreased by mnmzng the cost value whch s the fnal soluton n the form of: w Q, w Q, w 3Q M M M Q M (3) Q, Q, M M 3Q M x3 4. Characterstcs the Proposed HBMO In ths work, the proposed HBMO algorthm s used effectvely and effcently to desgn a front-end amplfer. The features of the HBMO verson can brefly be summarzed as follows: Our HBMO algorthm works effectvely wth a sngle Queen called as the Master Queen and generated randomly between predefned upper and lower lmtatons together wth her versatle genetc nhertance and has the duty of gve brth to the members of the N hve hves of the colony nstead of one hve. The rest of Queen Populaton s used for regstraton of the Queens of the N hve hves from where the Master Queen Q M wll be selected. Dvson of the entre colony nto the N hve hves facltates ortng process appled to the sub-colones step by step, n the other words the search for the new Canddates s performed n a reduced number of sub-matrces nstead of makng a search for a sngle ggantc matrx. Ths gans the algorthm both smplcty and effcency. The matng process s also smplfed to only energy-based probablstc decson rule to enable the more fttest solutons, and furthermore Egg populaton s defned as a random selecton process based on a crossng over to generate search space as a sub-space of the entre huge soluton space, besdes Royal Jelly feed s used n algorthm to make a local search n order to mprove the ftness of the Master Queen bee at the end of the each generaton or teraton. Thus the comparson wth the counterpart populaton-based algorthms verfed that a robust and fast convergent algorthm wth the mnmal problem nformaton s resulted for the desgn of a front-end amplfer. 5. Worked Example The user-defned parameters of the HBMO algorthms are set to the followng values n the desgn of the frontend amplfer: N Drone =, t max = N Gen =, m Q =, m D =, N Rj =, E max =, E mn =., cost req =.. In the work example, NE35 s selected as the mcrowave transstor and ts mnmum nose fgure F mn (f) profles are gven n Fg. 5 at several bas condtons.

7 4 F. GÜNEŞ,. DEMİRE, P. MAHOUTİ, DEIGN OF A FRONT END AMPIFIER FOR THE MAXIMUM POWER DEIVERY. F mn (db).6.4 V D =V, V D =V, V D =V, V D =3V, I D = 5mA I D =ma I D =ma I D =ma Fg. 5. Nose Fgure Profles at several bas condtons. Furthermore, the maxmum gan G Tmax (f) varatons constraned by the mnmum nose fgures F mn (f) and F = db are evaluated numercally usng the HBMO and compared the analytcal counterparts [], [] whch are taken place n Fgs. 6 and 7 at several bas condtons, respectvely. 6 Fg. 8. NA wth T- type Mcrostrp Matchng Networks. W (mm) W (mm) W 3 (mm) W 4 (mm) W 5 (mm) W 6 (mm) l (mm) l (mm) l 3 (mm) l 4 (mm) l 5 (mm) l 6 (mm) Tab.. olutons of the T type nput and output mcrostrp matchng elements for the maxmum output power and the nose fgure F(f) = db. Impedance msmatchng at the nput and output ports are gven as compared wth the Genetc Algorthm (GA), Partcle warm Optmzaton (PO) and HBMO wth and wthout Royal Jelly n Fgs. 9 and, respectvely. 5 4 V D =V, V D =V, V D =V, I D = 5mA I D =ma I D =ma.95 G Tmax (db) 3 9 V D =3V, I D =ma M n.9 5 GA PO HBMO Fg. 6. Constraned Maxmum gan G Tmax (f) by F mn (f). G Tmax (db) M out Fg. 9. Comparson of algorthm for mpedance msmatchng at the nput port F req = F mn F req = db F req =.5 db GA PO HBMO Fg. 7. Constraned Maxmum Gan by F req F mn frequency characterstcs at the bas condton (V D = V, I D = ma). The gan performance G Tmax (f) constraned by F = db at the bas condton ( V, ma) s desgned on the substrate of Rogers 435 (ε r = 3.48, h =.54 mm, tanδ =.3, t =. mm) along the bandwdth of 5- GHz. The soluton spaces of the T-type matchng crcuts n Fg. 8 are shown n Tab Fg.. Comparson of algorthm for mpedance msmatchng at the output port. The resulted gan, nose performances, nput and output reflectons of the amplfer desgned by HBMO wth Royal Jelly are gven n Fgs.,, 3, 4, respectvely as compared wth the targeted performances and obtaned by the AWR crcut and 3-D EM smulators.

8 RADIOENGINEERING, VO. 3, NO., APRI 4 4 G T (db) Nose Fgure (db) Target EC mulaton EM mulaton Fg.. Comparatve gan performance of the amplfer for the maxmum power delvery and the nose fgure of F(f) = db EC mulaton Fg.. yntheszed nose performance of the T-type amplfer. Input Reflecton EC mulaton EM mulaton Fg. 3. Input reflecton of the T-type amplfer. Output Reflecton Target EC mulaton EM mulaton Fg. 4. Output reflecton of the T-type amplfer. Furthermore the cost and executon tme wth teraton number of the used counterpart s algorthms whch are GA, PO, and HBMO wth and wthout Royal Jelly are gven n Fg. 5. Cost 5 5 Cost PO Cost GA Cost HBMO Cost Tme PO Tme GA Tme HBMO Tme Iteraton Fg. 5. Cost and executon tme varatons for PO, GA and. The optmzaton parameters of the studed algorthms are gven n Tab. 5, the parameters of the PO and GA are taken as ther default values of the MATAB Optmzaton tool, MATAB b []. The cost values and executon tmes at the th teraton of a random run are gven n Tab. 3 performed by the Intel Core 7 CPU,.6 GHz Processor, 6 GB RAM. Furthermore the statstcal analyss s performed benchmarkng of the selected algorthms for tmes of tres depcted n Tab. 4 that result n a hgh success rate of the proposed algorthm. Thus one can nfer the effectveness and effcency of the proposed HBMO based desgn by comparng t wth the GA, PO and the smple HBMO based desgns. Algorthm Cost Executon Tme (ec) HBMO &Royal Jelly.7 84 HBMO.77 7 PO.5 84 GA.5 89 Tab. 3. Benchmarkng at th teraton. Algorthm Worst Best Mean HBMO &Royal Jelly.9. HBMO GA PO Tab. 4. Benchmarkng of cost varaton for tres at th teraton for all algorthms. Algorthm HBMO& Royal Jelly Populaton Maxmum Iteraton Iteraton 5 5 HBMO Iteraton Executon Tme (ec) pecal Parameters N Drone =, E max =, E mn =., N RJ tep ze = ±. N Drone =, E max =, E mn =. GA 3 5 Gaussan Mutaton PO 3 5 earnng factors c = c = Tab. 5. User defnes parameters of the algorthms.

9 4 F. GÜNEŞ,. DEMİRE, P. MAHOUTİ, DEIGN OF A FRONT END AMPIFIER FOR THE MAXIMUM POWER DEIVERY 6. Conclusons Ths work has gathered the analytcal and novel artfcal ntellgence technques developed by our research team [-9], to desgn a front-end amplfer that s formulated as an optmzaton problem each ngredent of whch s carred out rgorously on the mathematcal bass. In fact, ths front-end amplfer desgn s based on a constraned optmzaton problem wth the dfferent objectves whch are the maxmum sgnal power delvery and the requred nose. Besdes n ths work, the cost effectve 3-D ONNET-based VRM mcrostrp model s used as a fast and accurate model worked out by our research group n the desgn process of the mcrostrp matchng networks. Another orgnalty s buldng of a smple and effcent verson of the HBMO to be used n the front-end amplfer desgn. The HBMO combnes the powers of smulated annealng, genetc algorthms to search for the best possble soluton to the problem under nvestgaton wthn a reasonable computng tme. Furthermore, a smple local heurstc s combned wth the HBMO to ncrease the accuracy wthout decreasng much the computatonal effcency. Thus, the sgnfcance of the work for the mcrowave crcut theory can manly be temzed as follows: () Frst of all, the desgn needs solely the fundamental mcrowave crcut knowledge; () Desgn target s based on the potental performance of the used actve devce that s obtaned by solvng numercally the nonlnear gan, nose and nput and output msmatchng equaton usng the HBMO subject to the desgn objectve; () In the desgn of the nput and output mcrostrp matchng crcuts, the cost effectve mcrostrp VRM model s used as a fast and accurate model so that t facltates to obtan drectly all the matchng mcrostrp wdths, lengths { W, } on a chosen substrate to satsfy the maxmum power delvery and the requred nose over the requred bandwdth of a selected transstor; (v) Mcrostrp matchng crcut n any confguraton can be easly syntheszed by the HBMO wth the royal jelly fast and accurately compared to the other counterpart evolutonary algorthms. It can be concluded that the paper presents an attractve desgn method for a front-end amplfer desgn based on the transstor potental performance, and t can be adapted to desgn of the other types of lnear amplfers. References [] DEMIRE,. A generalzed procedure for desgn of mcrowave amplfers and ts applcatons. PhD Thess (n Turksh). Yıldız Techncal Unversty, Istanbul, Turkey, 9. [] DEMIRE,., GÜNEŞ, F. Performance characterzaton of a mcrowave transstor for maxmum output power and the requred nose. IET Crcuts Devces yst, 3, vol. 7, no., p. 9. [3] HADDAD, O. B., AFHAR, A., MARINO, M. A. Honey-Bees Matng Optmzaton (HBMO) algorthm: A new heurstc approach for water resources optmzaton. Water Resources Management, 6, vol., p [4] AFHAR, A., HADDAD, O. B., MARINO, M. A., ADAM, B. J. Honey-Bee Matng Optmzaton (HBMO) algorthm for optmal reservor operaton. Journal of the Frankln Insttute, 7, vol. 344, p [5] FATHIAN, M., AMIRI, B., MAROOI, A. Applcaton of Honey- Bee Matng Optmzaton algorthm on clusterng. Appled Mathematcs and Computaton, 7, vol. 9, p.5 3. [6] NIKNAM, T. Applcaton of Honey Bee Matng Optmzaton on dstrbuton state estmaton ncludng dstrbuted generators. J. Zhejang Unversty CIENCE A, 8, vol. 9, p [7] VAPNIK, V. The Nature of tatstcal earnng Theory. New York: prnger Verlag, 995. [8] KEKIN, A. K. Desgn optmzaton of ultra wde band mcrostrp amplfer usng 3-D onnet-based VRM wth partcle swarm ntellgence. Mc Thess. Yıldız Techncal Unversty, Istanbul, Turkey,. [9] [] POZAR, D. M. Mcrowave Engneerng. John Wley & ons,. [] GÜNEŞ, F., GÜNEŞ, M., FIDAN, M. M. Performance characterzaton of a mcrowave transstor. IEE Proceedngs- Crcuts, Devces and ystems, 994, vol. 4, no. 5, p [] GÜNEŞ, F., ÇETINER, B. A. A Novel mth chart formulaton of performance characterzaton for a mcrowave transstor. IEE Proceedngs -Crcuts Devces and ystems, 998, vol. 45, no. 6, p [3] GÜNEŞ, F., ÖZKAYA, U., DEMIRE,. Partcle swarm ntellgence appled to determnaton of the feasble desgn target for a low nose amplfer. Mcrowave and Optcal Technology etters, 9, vol. 5, p [4] MAHOUTI, P., GÜNEŞ, F., DEMIRE,. Honey Bees Matng Algorthm appled to feasble desgn target space for a wde band front end amplfer. In ICUWB - IEEE Internatonal Conference on Ultra-Wdeband. yracuse (NY),, p. 5 55, DOI:.9/ICUWB [5] GÜNEŞ, F., TÜRKER, N., GÜRGEN, F. gnal-nose support vector model of a mcrowave transstor. Int J RF and Mcrowave CAE, 7, vol. 7, no. 4, p [6] GÜNEŞ, F., TORPI, H., GÜRGEN, F. A multdmensonal sgnalnose neural network model for mcrowave transstors. IEE Proceedngs-Crcuts Devces and ystems, 998, vol. 45, no., p. 7. [7] DEMIRE,., GÜNEŞ, F., TORPI, H. Partcle swarm ntellgence use n feasble desgn target space of a mcrowave transstor for a wde-band output-stage requrements. In ICUWB- IEEE Internatonal Conference on Ultra-Wdeband. yracuse (NY),, p. 46-5, DOI:.9/ICUWB [8] GÜNEŞ, F., KEKIN, A. K., DEMIRE,. Genetc Algorthm appled to mcrostrp mplementaton of matchng crcuts for a UWB low-nose amplfer. In IEEE Internatonal Conference on Ultra-Wdeband. yracuse (NY),, p. 4 45, DOI:.9/ICUWB [9] GÜNEŞ, F., TOKAN, N. T., GÜRGEN, F. A knowledge-based support vector synthess of the transmsson lnes for use n mcrowave ntegrated crcuts. Expert ystems wth Applcatons,, vol. 37, no., p [] MATAB and Neural Networks Toolbox Release b, the MathWorks, Inc., Natck, Massachusetts, Unted tates.

10 RADIOENGINEERING, VO. 3, NO., APRI 4 43 About Authors... Flz GUNE receved her M.c. degree n Electroncs and Communcaton Engneerng from the Istanbul Techncal Unversty. he attaned her Ph.D. degree n Communcaton Engneerng from the Bradford Unversty n 979. he s currently a full professor n Yıldız Techncal Unversty. Her current research nterests are n the areas of multvarable network theory, devce modelng, computer aded mcrowave crcut desgn, monolthc mcrowave ntegrated crcuts, and antenna arrays. alh DEMIRE has receved M.c. and Ph.D. degrees n Electroncs and Communcaton Engneerng from Yıldız Techncal Unversty, Istanbul, Turkey n 6 and 9, respectvely. He has been currently workng as an assstant professor n the same department. Hs current research nterests are among of mcrowave crcuts especally optmzaton of mcrowave crcuts, broadband matchng crcuts, devce modelng, computer-aded crcut desgn, mcrowave amplfers. Peyman MAHOUTI receved hs M.c. degree n Electroncs and Communcaton Engneerng from the Yıldız Techncal Unversty n. He has been currently n Ph.D. program of Yıldız Techncal Unversty. The man research areas are optmzaton of mcrowave crcuts, broadband matchng crcuts, devce modelng, computeraded crcut desgn, mcrowave amplfers.