IJCSNS Internatonal Journal of Computer Scence and Network Securty, VOL.16 No.5, May 2016 45 A smulaton-based optmzaton of low nose amplfer desgn usng PSO algorthm Roohollah Nakhae, Peyman Almasnejad and Mohammad Zahab Department of electrcal Engneerng, Payame Noor Unversty, I.R. of Iran Department of computer Engneerng and nformaton technology, Payame Noor Unversty, I.R. of Iran Summary In ths paper we propose a partcle swarm optmzaton based approach for desgnng CMOS low nose amplfer (LNA). Exstence of tradeoffs between nose, gan, lnearty, stablty and power consumpton n LNA desgn, forces the desgner to accomplsh such a long tme complcated work to optmze the crcut. So the utlzaton of a modern optmzaton s nevtable. In ths work the usage of PSO algorthm n LNA crcut desgn optmzaton has been nvestgated. The performance of LNA crcut s evaluated through HSPICE and PSO algorthm s mplemented n MATLAB, so a combnaton of MATLAB and HSPICE s performed. A cascode LNA s desgned usng 0.18µm CMOS technology foe near 2.4 GHz band applcatons. The results of ths work ndcate the effectveness of ths optmzaton approach that s a tme saver method. Key words: LNA, PSO, optmzaton, CMOS. 1. Introducton Nowadays, due to emergng commercal wreless applcaton, hgh-performance RF crcuts are requred n modern communcaton. Low nose amplfer (LNA) s one of the most mportant and essental buldng blocks n RF transcevers [1]. It s ntegrated nto the recevng chan and s ether drectly connected to the antenna or placed after RF pass band flter[2], and t should provde a proper low nose amplfed sgnal to the next stage, e.g. mxer, from a week nput sgnal by addng as lttle nherent as possble[3]. The LNA should provde suffcent tranceconductance gan wth acceptable lnearty and power consumpton to allow for long battery lfe especally n portable hand-held applcatons. Therefore there are some tradeoffs between gan, nose fgure (NF), lnearty and stablty n LNA desgn optmzaton [4]. In order to acheve the best performance of the desgned crcut, desgners must contnuously and repeatedly tune the desgned crcut elements and perform a crcut smulaton usng an electrc computer-aded desgn (ECAD) software, to optmze actve devces model parameters and szes, passve devces parameters, basng condtons, etc. t s n general a long tme, hgh complexty and complcated work. A proper optmzaton method can be used to overcome ths problem. In ths work we mplement a partcle swarm optmzaton (PSO) algorthm n MATLAB [5] whch has been lnked wth an electrcal smulator, HSPICE [6]. The performance of LNA s consdered n terms of NF, IIP3, gan, s- parameters and power consumpton. Prce and other market requrements force RF recevers to be ntegrated n standard CMOS technology along wth the rest of dgtal sgnal processng unt [7]. Snce several applcatons have been developed near the 2.4 GHz band, such as IEEE802.15.4, IEEE802.11b and Bluetooth, we have smulated a 2.4 GHz CMOS LNA usng TSMC parameters for 0.18 μm mxed sgnal and BSIM30 verson 3.1[8]. Ths paper s organzed as follows: The partcle swarm optmzaton concept s explaned n Secton 2. In Secton 3 the platform of optmzaton s explaned. In Secton 4 the LNA desgn specfcatons and consderaton has been descrbed. We propose the method of optmzng LNA desgn usng PSO n Secton 5. The acheved smulaton results are dscussed n Secton 6 and fnally, we draw conclusons n secton 7. 2. Partcle swarm optmzaton Partcle swarm optmzaton (PSO), frst ntroduced by Kennedy and Ebehart[9], s an evolutonary computaton method based on the socal behavor and movement of swarm searchng for the optmal and best locaton n a multdmensonal search space and has been found to be robust n solvng contnues nonlnear optmzaton problems[10]. Ths approach smulates the socal behavor of brd flockng or fsh schoolng model. Each partcle (-th partcle) poston s represented by a d-dmensonal vector and denoted as X = [x1, x2,, xd ] and s randomly ntalzed. The set of n partcle n the swarm are called populaton: X=[X1, X2,,Xn]. Each partcle s assumed to move around n the so called multdmensonal space to reach the best poston whch has the best ftness value. In each teraton of smulaton the ftness functon s evaluated by takng the current poston of each partcle, Manuscrpt receved May 5, 2016 Manuscrpt revsed May 20, 2016
46 IJCSNS Internatonal Journal of Computer Scence and Network Securty, VOL.16 No.5, May 2016 and f acheved ftness value s better than prevous best ftness of -th partcle, the current poston wll be selected as the best prevous poston of -th partcle and descrbed as PB=[pb1, pb2,, pbd]. The best poston among the populaton s called global best poston and descrbed as GB=[gb1, gb2,, gbd]. The rate of poston change for each partcle s called partcle velocty: V = [v1v2,, vd ]. each partcle would lke to return to ts own optmum pont, so the velocty has a term proportonal to (pb-x), t would lke to follow overall best global optmum pont too, so a term proportonal to (gb-x) s added to velocty. Therefore: vdk+1 = wvdk +c1 rand1k (pbd -xk) + c2 rand2k (gbdk-xdk) (2) Where w s ntera weght parameter whch controls the tradeoff between the global and the local search capabltes of the swarm. c1 and c2 are acceleraton factors and ndcate the relatve attracton toward pb and gb respectvely. rand1 and rand2 are two random numbers unformly dstrbuted between 0 and 1, whch ndcate the crazness of partcles[10]. k s the teraton number. The new poston of -th partcle s then determned by: xdk+1 = xdk + vdk+1 (2) Generally PSO has the advantage of beng very smple n concept, easy to mplement and computatonally effcent algorthm. Snce updates n algorthm consst of smple addng and multplcaton operators and no dervaton operaton s ncluded, computaton tme s dramatcally decreased compared to other heurstc algorthms. In order to avod premature Convergence, PSO utlzes a dstnctve feature of controllng a balance between global and local exploraton of the search space whch prevents from beng stacked to local mnma [11]. 3. System platform The purpose of optmzaton s to acheve the optmal sze of MOS transstors (channel length and wdth), passve component values and bas currents of the crcut, n order to meet the desred specfcatons. Ths paper proposes a technque that utlzes a smulaton-based approach for crcut desgn optmzaton. The system platform s llustrated n Fg. 1. The startng pont s a spce-lke netlst of the crcut topology, currently entered n the optmzaton engne as ts nput. Desred specfcatons are other nputs. Optmzaton engne consst of a PSO algorthm wrtten n MATLAB whch has been lnked wth an electrcal smulator, HSPICE, as ts performance evaluator. Fg. 1 system platform flow. Optmzaton s executed on a vector of desgn varables of crcut extracted from the netlst. In fact, the partcle poston vector s the channel length (L) and wdth (W) of MOS transstors, passve component values, and bas currents. At frst, each partcle (-th partcle) poston vector s randomly ntalzed. In each teraton, the algorthm runs the HSPICE for each partcle and evaluates a cost functon (CF), descrbed n (3): k f ( x) f d (3) CF = w = 1 f d Whch, f(x) s -th desgn specfcaton, k s the number of specfcatons, fd s desred value for -th desgn specfcaton and w are the mportance factors that say whch specfcaton has more mportance for desgner regardng to dfferent especal applcatons of the crcut. Wth ths approach, the searchng algorthm (optmzer) wll fnd the best soluton that meets ths cost functon.
IJCSNS Internatonal Journal of Computer Scence and Network Securty, VOL.16 No.5, May 2016 47 Therefore, PB and GB are showng the mnmum CF for best prevous poston and global best poston respectvely that each partcle attempts to reach them. At the end of each teraton, GB gves the best soluton of the populaton. g do d( = d( V d DS ) ) V DS = 0 4. LNA desgn specfcatons As called prevously, there are some tradeoffs between nose fgures (NF), gan, lnearty and stablty n LNA desgn process. There are some consderatons about these desgn Specfcatons: 4.1 Lnearty Increasng the gan of LNA can degrade the lnearty. Lnearty of LNA s measured n terms of IIP3 that s requred to be maxmzed. Generally, to acheve hgher lnearty n RF recevers, IIP3 must be more than -10dBm [12]. 4.2 Stablty The stablty condtons, f and only f k>1 and Δ <1, presented by stablty factor k must be satsfed [3] where k= (1- S11 2- S22 2+ Δ ) / (2 S21 S12 ) (3) Δ= S11 S22 - S21 S12 (4) For CMOS LNAs, the requred S21 s normally larger than 10 db. For S11 and S22, the nput/output return losses, less than -10 db are desrable. Also to avod unwanted sgnals to reach the LNA nput from the followng stages, reverse power gan S12 must be less than -20 db [12].so n LNA desgn optmzaton these specfcatons must be consdered. 4.3 Nose The nose performance of a crcut s typcally characterzed by a nose factor F or NF whch ndcates how much degradaton occurs n the output sgnal-to-nose rato due to the crcut s nternal nose [13]: n total output nose F = = out total output nose due to source or n NF = 10 log out (5) Fg.2 shows the standard CMOS nose model [14]. One of the nose source caused by channel resstance, whch s modulated by Vgs. ts power spectral densty gven by: 2 =, (6) 4kTγ nd g do Fg. 2 Standard CMOS nose model. Whch, gdo s the conductance when VDS s equal to zero. γ s the channel thermal nose coeffcent, whch depends on channel length and ts bas condtons. Fnte gate resstance also exhbts the other source of nose. Power spectral densty of gate s thermal nose s gven by: v 2 = 4, (7) R ng ktr G RgW = 3n L G 2 Where Rg s the gate polyslcon sheet resstance, W and L are the wdth and length of the devce and n s the number of gate fngers n the devce layout. Therefore, sze and other parameters of actve devces and the value of crcut's resstance and bas condton affect the NF and must be optmzed n LNA desgn. 5. The LNA crcut desgn and optmzaton The schematc of LNA under nvestgaton [3] s shown n fg.3. It has the cascode topology wth nductve degeneraton to provde a hgh gan and hgh reverse solaton, whch mprove the stablty and smply nput port matchng. Fg. 3 The cascade LNA schematc.
48 IJCSNS Internatonal Journal of Computer Scence and Network Securty, VOL.16 No.5, May 2016 The purpose of LNA optmzaton s to acheve the actve devces sze and passve values that gve the best desgn specfcatons, whch consdered n secton 3.for ths purpose; the PSO algorthm gets a spce netlst of LNA crcut as ts nputs. In fact, the partcle poston vector s the sze and value of the actve and passve devces. In ths work, t s a 9-dmensonal vector, as shown n table 1. Table 1: The desgn parameters and specfcatons The poston vector's elements Desgn specfcatons Iref NF WM1 IIP3 WM2 Power dsspated WM3 S11 (nput return l ) Ld S22 (output return l ) Ls S21(power gan) Lg S12(reverse power gan) C Rb At the optmzaton engne, each partcle (-th partcle) selects random values for actve and passve devces (or random value for -th poston vector). In each teraton, the algorthm runs the HSPICE for each partcle and evaluates a cost functon (CF), descrbed n (3). So, at the end of each teraton, PB and GB are showng the mnmum CF for best prevous poston and global best poston respectvely that each partcle attempts to reach them. Fnally, as the stop condton of algorthm s reached, GB shows the optmzed values of the crcut. In case study 1, the gan has assumed to get much hgher prorty among other desgn specfcatons. As ndcated n table 3, the mprovement n gan s acheved n cost of lnearty degradaton and power dsspaton. Conversely, the power has more mportance for desgner, n case study 2. It s clearly vsble from table 3 that although the power consumpton s decreased but the gan of LNA s decreased too. Fnally, consderng all specfcatons, a reasonable LNA desgn for general applcatons s carred out n case study 3, whch ts gan, s11, s12 and NF are depcted n fg.4 6. Smulaton results Consderng so called mportance factors, three case studes of LNA s carred out usng PSO. For each case, acheved value of passve devces and sze of actve devces and bas current are tabulated n table 2, and a comparson wht three other works s shown n table 3. Table 2: acheved value of passve devces and sze of actve devces and bas current Fg. 4 The specfcatons of case study 3. Table 3: The crcut specfcatons n comparson wht other works
IJCSNS Internatonal Journal of Computer Scence and Network Securty, VOL.16 No.5, May 2016 49 7. Concluson In ths paper, a PSO based approach for optmal desgn of LNA crcut has been reported. Electrcal characterstcs of the LNA crcut consdered n the optmzaton process are the gan, S parameters, nose fgure, power consumpton and the nput thrd-order ntercept pont. The results of ths work ndcate the effectveness of ths optmzaton approach that s a tme consumng method. We note that ths approach can also be appled to desgn optmzaton of other crcut and can be embedded nto any electronc CAD software whch mproves the process of desgn and fabrcaton. Transactonson Mcrowave Theory and Technques (2006) 4062 4071. [17] Muhammad Khorram and S.M. Resaul Hasan A 3 5 GHz Current-Reuse gm-boosted CG LNA for Ultrawdeband n 130 nmcmos, IEEE TRANSACTIONS ON VERY LARGE SCALE INTEGRATION (VLSI) SYSTEMS, VOL. 20, NO. 3, MARCH 2012, pp-400-409. [18] S.Udaya shankar, M.Davdson Kamala dhas, " Desgn and Performance Measure of 5.4 GHZ CMOS Low Nose Amplfer usng Current Reuse Technque n 0.18.m Technology", Proceda Computer Scence 47 ( 2015 ) 135 143 References [1] A. Tell, M. Askar, CMOS LNA desgn for LEO space S- band applcatons, IEEE Can. Conf. Electr. Comput. Eng. 1, 2003 )27 30). [2] B. Razav, RF Mcroelectroncs, Prentce Hall Inc., 1998R [3] S.Toofan et al, A low-power and hgh-gan fully ntegrated CMOS LNA, Mcroelectroncs Journal 38 (2007) 1150 1155 [4] S.Park, W.Km, Desgn of a 1.8GHz low-nose amplfer for RF front-end n a 0.8 mm CMOS technology, IEEETrans.Consum.Electron.47 (1)(2001). [5] http://www.mathworks.com/products/matlab [6] http://www.synopsys.com/products/mxedsgnal/hspce/hsp ce.htm [7] J.-H.Tsa, W.-C.Chen, T.-P.Wang, T.-W.Huang, H.Wang, Amnature Q-band low nose amplfer usng 0.13 mm CMOS Technology, IEEE Mcrowave Wreless Components Lett.16(6)(2006)327 329 [8] http://www.moss.com [9] J. Kennedy, R.C. Eberhart, Partcle swarm optmzaton, n: Proc. IEEE Internatonal Conference on Neural Networks, 1995, pp. 1942 1948. [10] J.schneder, S.krkpatrck, Stochastc optmzaton, sprnger-verlag berln Hedelborg,2006. [11] ] M. Clerc, The partcle swarm Exploson, stablty and convergence n a multdmensonal complex space, IEEE Trans. Evol. Comput. (2002) 58 73. [12] S. Park, W. Km, Desgn of a 1.8GHz low-nose amplfer for RF front-end n a 0.8 mm CMOS Technology, IEEE Trans. Consum. Electron. 47 (1) (2001) 10 15. [13] S. Toofan et al, Low power and hgh gan current reuse LNA wth modfed nput matchng and nter-stage nductors, Mcroelectroncs Journal 39 (2008) 1534 1537. [14] D.K. Shaeffer, T.H. Lee, A 1.5-V, 1.5-GHz CMOS low nose amplfer, IEEE J. Sold-State Crcut 32 (5) (1997) 745 759 [15] L.-H. Lu, H.-H. Hseh, Y.-S. Wang, A compact 2.4/5.2-GHz CMOS dual-band low-nose amplfer, n: IEEE Mcrowave and Wreless Components Letters, vol. 15, 10, 2005, pp. 685 687. [16] T.-K.Nguyen,V.Krzhanovsk,J.Lee,S.-K.Han,S.-G.Lee,N.- S.Km,C.-S.Pyo, A low-power RF drect-converson recever/transmtter for2.4-ghz-band IEEE 802.15.4 standard n 0.18-mm CMOS technology, IEEE