Noise Figure Evaluation Using Low Cost BIST

Transcription

1 Noise Figure Evlution Using Low Cost BIST Mrelo Negreiros, Luigi Crro, Altmiro A. Susin Instituto de Informáti - Progrm de Pós Grdução em Computção - PPGC Universidde Federl do Rio Grnde do Sul - UFRGS, Porto Alegre, RS, Brzil negreiro,rro,susin@inf.ufrgs.br Abstrt A tenique for evluting noise figure suitble for BIST implementtion is desribed. It is bsed on low ost single-bit digitizer, wi llows te simultneous evlution of noise figure in severl test points of te nlog iruit. Te metod is lso ble to benefit from SoC resoures, like memory nd proessing power. Teoretil bkground nd experimentl results re presented in order to demonstrte te fesibility of te ppro. 1. Introdution Wit te inresing pity of development ieved by te eletroni industry nowdys, minly in te onsumer mrket, te design of new devies is gret llenge. A fst, flexible nd relible metodology must be used in order to ope wit ever srinking produt development time, s sorter time-to-mrket is ruil to te ommeril suess of su devies. Te SoC ppro [1] s been used in order to solve tis issue. Testing of su devies is noter llenge, s test osts must be kept lower for te devie to be ompetitive in te mrket. Unfortuntely, nlog testers re expensive devies nd extr diffiulties rise wen trying to test nlog iruits in SoC environment. Te limited ess to te input nd output of te nlog iruit under test is n exmple. In order to redue te ost of te nlog test, built-in test strtegies my be used in order to redue te requirements of externl nlog testers, or reple tem ompletely. In te SoC environment, s plenty of proessing nd memory resoures re vilble, it is possible to perform test nlysis by reusing tese resoures [2]. In tis pper low ost metod for evluting te noise figure of nlog iruits is presented. Noise figure is n importnt prmeter in te speifition nd design of low noise systems, su s ommunitions systems nd biomedil instrumenttion. It is used to rterize te noise bevior, from single nlog omponents to entire systems. Te test metod is bsed on digitl signl proessing nd my be implemented in te SoC environment by reusing proessing nd memory resoures lredy vilble in te SoC. A one-bit digitizer tt is permnently onneted to te desired nlog test point is used, tus minimlly disturbing te iruit under test. Tnks to te simpliity of te onverter, low nlog re overed is obtined, nd no impt is mde on te noise figure of te iruit being tested, tt is, te proposed BIST does not inrese te noise level to be mesured. Te ultimte gol of tis work is to sow tt, by using simple BIST ell [3], one n mesure not only frequeny relted prmeters of te iruit under test, but rter one n obtin informtion of oter importnt rteristis like noise figure. Te pper is orgnized s follows: in setion 2 brief review of oter pproes to mesure te Noise Figure is presented. In setion 3, noise prmeters re reviewed nd two ommon noise figure mesurement metods re presented. An nlysis of possible implementtion of te metods ere proposed in te SoC environment is provided in setion 4, togeter wit te test metod. In setion 5 results regrding te digitizer re presented, followed by experimentl noise figure evlutions. Anlysis is provided in setion 6 nd te pper finises wit onlusions in setion Relted work Te use of embedded noise soures for noise figure mesurements s lredy been reported. In [4] tinfilm resistor s been used s n on-wfer noise soure. Diode noise soures ve been investigted in [5]. A prodution test seme bsed on signture testing s been proposed in [7], were noise figure mesurements ve been indiretly mde. To te utors knowledge, first ppro sowing tt it is possible to ve BIST iruit to mesure noise figure s been proposed in [6]. In te present work we ddress te issue of using low /5 $2. 25 IEEE

2 ost BIST to mesure te noise figure. 3. Noise in nlog iruits In tis setion some bkground informtion regrding noise rteriztion in nlog iruits is provided. First noise figure is defined, nd ten two usul mesurement metods re presented. 3.1 Definitions A ommon prmeter used to rterize te noise bevior of n nlog eletri signl (su s te output of sensor or mplifier) is te signl-to-noise rtio (SNR). It is rtio of te signl power to te noise power, expressed in db. 2 V 1 log1 db 2 V S SNR = (1) N Noise figure (NF) nd noise ftor (F) re prmeters used to rterize te noise bevior of devie or iruit, s sown in figure 1. SNR IN R S SNR OUT Figure 1. Noise rteriztion of 2-port devie Te noise ftor (F) of two-port devie is te rtio of te vilble output noise power per unit bndwidt to te noise used by te tul soure onneted to te input terminls of te devie, mesured t te stndrd temperture of 29K [8]. SNR SNR IN (2) OUT Noise figure (NF) is defined s te noise ftor (F) expressed in db [8,9] N + k T (4) k T Te definition of noise figure is bsed on te ssumption of liner system. Some extensions for nonliner systems ve been proposed, like in [1], but re not going to be nlyzed in tis work. Some usul noise figure vlues re illustrted in tble 1, togeter wit te orresponding noise ftor. Note tt te typil vlue of noise figure for n RF low noise mplifier is 3dB. For n RF mixer, te vlue is bout 1 db [11]. A iruit tt does not dd noise to its input would ve noise figure of db. Tble 1. Some referene vlues for noise figure nd noise ftor NF(dB) F Exmple 1 noiseless nlog iruit 3 2 RF low noise mplifier 1 1 RF mixer 3.2 Mesurement metods In te diret mesurement metod, eqution 4 is used. A lod is onneted to te input of te system, t temperture of 29K, like in figure 1. Te output noise power of te system is mesured (te numertor of eqution 4). If one knows te mesurement bndwidt (B) nd te gin of te system (G), eqution 4 n be used diretly. In te y-ftor metod, noise figure evlution is bsed on te use of librted noise soure [8,12]. Te metod is two-step proess (see figure 2): wit te noise soure turned off (t temperture of 29K, or old temperture), te output power (N ) is mesured. Ten te noise genertor is turned on, nd te noise output power (N ) for te ot temperture is reorded. Te Y ftor is te rtio of tese powers: N N Y = (5) NF 1 log1 ( F ) db = (3) Te IEEE stndrd definition of te noise ftor (F) is given by eqution 4, were N is te noise dded by system, T is 29K (stndrd temperture), B is te system bndwidt, k is te Boltzmnn onstnt nd G is te gin of te system. Figure 2. Noise rteriztion setup Knowing tt te noise power t te output of te is simply te noise dded by te system (N ) plus te mplified input noise, one n write

3 N = k T + N (6) N = k T + N (7) After pplying equtions 6 nd 7 to 5, nd developing using eqution 4, one obtins te eqution of te Y-Ftor tenique [12], tt llows te evlution of te noise ftor ( T / T 1) Y ( T / T 1) ( Y 1) (8) T is te referene temperture of 29K. If te noise soure old temperture is not 29K, tis provides orretion term. Tis eqution n be rewritten in order to tke into ount noise powers, insted of tempertures [1], s sown in eqution 9. ( N / N 1) Y ( N / N 1) ( Y 1) (9) 4. A NF BIST in te SoC environment In tis setion te implementtion of metods for estimting noise figure suitble for BIST in SoC environment re disussed. Te proposed metod is lso presented. 4.1 Diret metod An implementtion of te diret metod would be s sown in figure 3. A nominl lod R s must be pplied t te input t temperture of T =29K. Te output of te must be mplified nd routed to te ADC of te system for furter proessing. R S T o G du G AD Figure 3. Diret metod setup One prtil disdvntge of tis setup is relted to vritions in te gin of te mplifier (G ). Tis gin multiplies bot terms in eqution 1, but only te numertor is mesured. Tis wy, ny devition in te mplifier gin (from G to G ) will use n error in te noise ftor estimtion. Tis issue is expeted to our beuse of proess vritions tt my ffet te gin of te mplifier. ( N + k To ) G ' k T G 4.2 Y-ftor metod o (1) If one ould embed suitble noise genertor, being ble to provide two known noise levels, it would be fesible to implement noise figure mesuring system bsed on te Y-ftor tenique. Te system level setup is presented in figure 4. A progrmmble ttenutor provides te noise levels needed for te NF mesurement. Te genertor noise level n be mesured troug n uxiliry nlog pt to te ADC. noise genertor T,T ttenutor AD G dut G Figure 4. Y-ftor setup Tis setup does not possess te sme sensitivity to nges in te mplifier gin s te diret metod. Bot numertor nd denomintor in eqution 11 re mesured, so ny devitions in te mplifier gin (from G to G ') re orreted. ( N + k T ) G ' Y = (11) ( N + k T ) G ' In [6] n nlysis of unertinty in ommeril noise soures is provided, nd it is sown tt even lrge errors like 5% in te ot temperture n still provide useful mesurements for noise figure estimtion, if n error of ±.3dB is eptble (for noise figures of 3dB nd 1dB). Tese errors ould be used s guidelines in te design of te noise genertor nd ttennutor. 4.3 Proposed metod for NF mesurement Some problems re not ddressed in te strtegy presented in figure 4: te AD onverter of te system is used, so simultneous quisition is not possible. Also, routing of nlog signls to te ADC my be diffiult. Tere is need for multiplexing devie t te input of te ADC, wi introdues non-linerity nd distortion in te signl. If te ADC is repled by simple digitizer, like in figure 5, some dvntges like te possibility of simultneous observbility nd no need for multiplexing devies is ieved. Also, beuse of te simpliity of te digitizer, it n be permnently onneted to te nlog test point, tus voiding swites wi degrde te performne of te nlog iruit under test.

4 noise genertor ttenutor mplifier referene wveform Figure 5. Proposed setup Digitizer As te Y-ftor metod requires te evlution of rtio of signl power, te digitizer must be ble to provide noise power levels. In tis work te digitizer [3] uses voltge omprtor nd referene signl in order to perform te dt quisition. Te requirements of te referene signl re modest (in te sense tt only smll frequeny bnd is used in te librtion proess), llowing simple nd low-ost signl genertor to be used. 5. Results In tis setion te digitizer nd strtegy to evlute power levels using referene wveform re presented. Noise nd referene levels re lso disussed. Te setion finises wit results from prototyped setup. 5.1 Digitizer Te digitizer (figure 6) is omposed by voltge omprtor wit noise referene [3]. Te input signl is onneted diretly to te input of te omprtor. Te digitized input signl is obtined t te output of te omprtor nd is digitl output. Smpling my be ontrolled by dding flip-flop t te output of te omprtor. Tis struture is ommon in ig speed voltge omprtors. Eqution 12 llows one to stte tt te sttistis of te input signl will be t te output of te smpler, ffeted by gin ftor nd by te rsine funtion, wi is pproximtely liner for smll vlues of te input rgument. As te Fourier trnsform of te utoorreltion is te power spetrum density, one is ble to observe te spetrl rteristis of te signl, but wit n inresed noise level beuse of te ddition of te noise t te omprtor input [3]. 5.2 Evluting power levels In te following, Mtlb simultion illustrtes te ide of mesuring noise levels using referene wveform. In rel pplition, noise nd referene wveforms sould be mplified in order to enble te use of voltge omprtor. If one pplies onstnt-mplitude squre wve signl to te digitizer of figure 6 s referene signl, noise levels n be determined if simple strtegy is followed. Two noise levels were pplied to te digitizer using te sme squre wve s referene. Signls re s sown in figure 7. One sould notie tt noise levels sould be lwys greter tn te referene levels. Figure 7. Noise nd referene wveforms for ot (left) nd old (rigt) noise tempertures signl noise + - D Q output smple Figure 6. Digrm of 1-bit digitizer Te noise mplitude sould be greter tn or equl to te signl mplitude, nd bot signls sould be zero men (or ve te sme d level). Te utoorreltion t te output of te smpler, onsidering tt te ombined signl nd noise is norml sttionry proess wit zero men, is given by te rsine lw: 2 Rx ( τ ) R = rsin y ( τ ) (12) π Rx () Figure 8. Power spetrum density Te power spetrum density evlution of te bitstrem t te output of te digitizer is sown in figure 8. One n notie tt te noise levels remins similr, wile mplitude levels of te referene squre wve re lrger. As te referene level is onstnt, simple normliztion proedure n be used. One n evlute te mximum mplitude of bot spetr nd pply

5 orretion ftor to one of te power spetrl density plots. error in power rtio [%] referene mplitude [Vref/Vnoise*1%] Figure 1. Error in power rtio estimtes versus referene mplitude 5.4 Experimentl results Figure 9. Power spetrum density fter normliztion (zoom t 6 Hz) Figure 9 sows different noise levels, obtined before nd fter te normliztion proedure. One n observe tt te noise levels were very lose before te normliztion proedure. In order to mke numeri omprison, noise power rtio ws evluted using tree different metods: rtio of men squre vlues (evluted in time domin), rtio of PSD dt nd rtio of PSD dt from te 1-bit digitizer. Te vlues obtined re presented in tble 2. For te 1-bit dt, te referene wveform must be exluded from te power rtio evlution (te referene is not prt of te signl being mesured). If tis is omplised, bout 2.5% error in te power rtio ws observed in te simultion, s presented by te lst line in tble 2. Tble 2. Noise power rtio evlution nd derived prmeters for T =1K nd T =1K Metod Noise power F NF(dB) rtio Men squre rtio PSD rtio bit PSD rtio exluding referene In order to verify te ppro test setup ws implemented in order to mesure te noise figure of non-inverting mplifier. Te generl setup is sown in figure 11. In order to nge te vlue of te noise figure of te iruit, different opertionl mplifier ws used. As te equivlent noise voltges re provided by te dtseets of te omponents, one is ble to lulte te expeted nominl vlue of te noise figure of te iruit, ording to te opmp used [13]. noise genertor ttenutor Non-Inverter Amplifier A v =11 Amplifier A v =1156 Sine genertor Voltge Comprtor Digitl output 3kHz 3mVpp Figure 11. Digrm of te experimentl setup Te opmps used were te OP27, OP7, TL81 nd CA314. Te orresponding expeted noise figures were in te rnge of 3.7dB to 16.2dB. In order to void interferene pikup, te iruit ws ssembled in n luminum se, being bttery powered (figure 12). 5.3 Noise nd referene levels Simultions were rried out in order to evlute te ury of noise power rtio estimtes s funtion of te mplitude of te referene wveform. Figure 1 sows te error in power rtio estimtes for gussin noise. Te referene wveform mplitude level is relted to te overll ury of te metod: for very smll mplitude referenes, lrge error is expeted beuse of noise levels disturbing te referene mplitude. Very lrge referenes my led to non-liner distortion of te digitizer. Amplitudes in te rnge of 1% to 4% of te noise level sould give resonble results (figure 1). Figure 12. Prototyped iruit Externl noise genertor nd sine wve genertor (bot HP3312A) were used. Te referene wveform ws t 3kHz, wile te noise mesurement bndwidt ws t 1kHz, s indited in figure 13. Te output of te digitizer ws quired using digitl sope (HP54645D). Dt ws proessed using Mtlb. Totl quisition lengt ws 1e6 smples nd te FFT size ws 1e4 smples. Te results obtined fter proessing re presented in tble 3,

6 inluding te expeted noise figure vlues from noise iruit nlysis [13]. 7. Conlusions In tis pper tenique for te evlution of noise figure in BIST environment ws presented. As te tenique is bsed on DSP, it is ble to benefit from resoures lredy vilble in te SoC environment. A simple voltge omprtor nd referene signl were used s digitizers, tus disrding te need for n ADC. Te tenique lso extends te pbilities of simple BIST ell [3], llowing one to perform frequeny nd noise mesurements. 8. Referenes Figure 13. PSD plot for noise levels fter normliztion Tble 3. Noise figure results for T =29K nd T =29K Opmp Expeted Mesured OP OP TL CA Anlysis Te experimentl results presented in setion 5 ve indited te fesibility of implementing te ppro for prtil iruits. Noise figure mesurements were rried out wit 2 db mximum bsolute error. Te proposed strtegy mkes use of referene signl in order to perform te normliztion proess. Even low-ost genertor ould be used, s te normliztion proess would trk te min frequeny omponent (disregrding rmonis, for exmple). Tis would enble te use of low qulity referene wveforms, s te rmonis re not used in te normliztion proess. Te mplitude of te min omponent, owever, sould be onstnt. Te need for nlog signl onditioning, s ig gin mplifier, is relted to te signl levels tt sould be mesured. In generl, n mplifier will be required for noise mesurements [9]. Tis overed ould be minimized by observing tt te noise figure of sde of stges is minly te noise figure of te first stge [9]. [1] Zorin,Y.; Mrinissen, E.J.,"System Cip Test - How Will It Impt Your Design", Design Automtion Conferene - DAC, 2, pp [2] Zorin,Y.; "Test resoure prtitioning strtegies for systems on ip", Ltin Amerin Test Worksop LATW23, TTEP Tutoril. [3] Negreiros, M.; Crro, L.; Susin, A.A., A Sttistil Smpler for New On-Line Anlog Test Metod, JETTA-Journl of Eletroni Testing: Teory nd Applitions 19, 23, Kluwer Ademi Publisers, pp [4] Belnd, P.; Roy, L.; Lbonte, S.; Stubbs, M.; "An enned on-wfer millimeter-wve noise prmeter mesurement system", IEEE Trns. on Instrumenttion nd Mesurement, vol.48, no.4, Aug. 1999, pp [5] J. Rnd, Robert L. Billinger, Jon L. Rie; On-Wfer Mesurements of Noise Temperture, IEEE Trns. on Instrumenttion nd Mesurement, vol.48, no.6, Deember 1999, pp [6] Negreiros, M.; Crro,L.; Susin,A.A.; Towrds BIST Tenique for Noise Figure Evlution, Europen Test Symposium, ETS 4, 24. [7] Voorkrnm, R.; Cerubl, S.; Ctterjee, A.; "A signture test frmework for rpid prodution testing of RF iruits", Design, Automtion nd Test in Europe Conferene nd Exibition, 22, pp [8] Fundmentls of RF nd mirowve noise figure mesurements, Hewlett-Pkrd Applition Note 57-1, Plo Alto, CA, July [9] Motenber, C.D.; Connelly, J.A.; "Low-Noise Eletroni System Design", Wiley, [1] Geens, A.; Rolin, Y.; "Noise figure mesurements on nonliner devies", IEEE Trns. on Instrumenttion nd Mesurement, vol. 5, n.4, Aug. 21, pp [11] Rzvi, B.; "RF Miroeletronis", Prentie Hll, [12] Collntes, J.-M.; Pollrd, R.D.; Syed, M.; "Effets of mismt on te noise figure rteriztion: omprtive nlysis of two Y-ftor teniques", IEEE Trns. on Instrumenttion nd Mesurement, vol.51, no.6, Deember 22, pp [13] Steffes, M.; "Noise Anlysis for Hig Speed Op Amps", Burr-Brown Applition Bulletin AB-13, Otober, 1996.