3 J.-M. WU, C.-K. LIOU, C.-J. CHUANG, Y. KUO, A. V T NETWORK-BASED. GHZ DOWNCONVERTER RFIC A. V T Netwok-Based. GHz Downconvete RFIC Jian-Ming WU, Ching-Kuo LIOU, Ching-Jui CHUANG, Yujin KUO Dept. of Electonic Engineeing, National Kaohsiung Nomal Univesity, Kaohsiung, Taiwan R&D Cente, Communication and Visual Poducts, HW RF Sec., Pegaton & Unihan Copoation, Taipei, Taiwan jianmingwu@nknucc.nknu.edu.tw Abstact. A. GHz downconvete RFIC is designed and implemented using a. μm CMOS standad pocess. An impotant goal of the design is to achieve the high lineaity that is equied in WiMAX systems with a low supply voltage. A passive T phase-shift netwok is used as an RF input stage in a Gilbet cell to educe supply voltage. A single supply voltage of. V is used with a powe consumption of 5.7 mw. The T netwok-based downconvete achieves a convesion gain (CG) of 5 db, a singlesideband noise figue (NF) of. db, an RF-to-IF isolation of geate than db, and an input-efeed thid-ode intecept point (IIP 3 ) of dbm when the LO powe of -3 dbm is applied. Keywods Downconvete, mixe, low voltage.. Intoduction Beyond 3G (B3G) systems, including the woldwide inteopeability fo micowave access (WiMAX) and the long tem evolution (LTE) systems, featue with high data ates, high mobility, and lage coveage aeas. An RF tansceive in B3G systems must have a low powe consumption to suppot long standby times of mobile devices. In a eceive, a downconvete is pimaily esponsible fo conveting fequencies. Opeating a downconvete at a low supply voltage can geatly educe the powe consumed by B3G eceives. Poposed techniques to design a downconvete with a supply voltage below V include: ) cuent euse []-[], ) the body-input scheme [5]-[7], and 3) use of a tansfome []-[]. The cuent euse technique povides high convesion gain (CG) but at the cost of a poo noise figue (NF). The body-input technique suppots tansisto bias voltages that ae lowe than the theshold voltage but a high LO powe (P LO ) is equied to dive the tansistos. The tansfome technique povides a elatively high isolation but a lage chip aea is needed. The goal of this wok is to design and implement a low supply-voltage downconvete with a low input LO powe using the TSMC. μm CMOS standad pocess fo WiMAX applications. A passive T-type o phase shifte that consists of two seies inductos and one paallel capacito is used as a single-ended-to-diffeential convete of an input RF signal. The tansistos in an active tansconductance stage of a commonly used Gilbet mixe ae absent, so the supply voltage of the downconvete can be educed. A downconvete in a WiMAX eceive is often equied to meet the high lineaity specification fo stictly high dynamic ange. Theefoe, the lineaity pefomance of the T netwok-based downconvete with an inducto load and a π-esonato load ae also compaed with each othe in this pape.. Cicuit Design Fig. schematically depicts the cicuit of the poposed downconvete with a T-type o phase-shift netwok. The T phase-shift netwok compises two seies inductances (L T ) and one shunt capacitance (C T ). The passive T netwok acts as a single-ended-to-diffeential convete. At fist, the single-ended RF input signal that is applied to this T phase-shift netwok geneates the diffeential RF signal. The tansistos M and M fom the two cuent souces, which have high output impedances to pevent the diffeential RF signal fom leaking to gound. The tansistos M 3, M, M 5, and M compise the multiplication coe that is diven by the local oscillato. Since this multiplication coe povides most of the convesion gain of the T netwok-based downconvete, the tansistos fom M 3 to M must be biased at a point in the satuation egion. The diect cuent (dc) bias design of the tansistos fom M 3 to M is based on the pinciples of a multiplie, so the tansistos ae not pushed to the cutoff state. Theefoe, the applied LO powe can be kept small in the downconvete. Then, the RF and LO signals ae mixed to geneate the output diffeential IF signal. To educe futhe the supply voltage, the inducto (L D ) is used as an output load because no dc voltage dop exists acoss it. Finally, the output voltage swing of the diffeential IF signal dops on the inductive load. As indicated above, the single-ended-to-diffeential convete is a T phase-shift netwok. The amplitude and phase diffeences between nodes RF + and RF, pesented in Fig., ae aimed to db and o, espectively. Notably, a capacitance value is much easie to implement in a CMOS standad pocess than an inductance value. Thee-
db o Degee RADIOENGINEERING, VOL., NO., APRIL 3 foe, the use of L T to detemine the C T in a T netwok design is pefeed. RF, LO, and IF fequencies of,, and 37 MHz, espectively, ae used in the downconvete. The applied LO powe is -3 dbm. Fig. plots the simulated amplitude and phase eos between nodes RF + and RF within the ange of C T values fom to. pf fo a selected L T value of.3 nh. Notice that the amplitude and phase eos ae the deviation of the amplitude and phase diffeences fom db and o, espectively. They ae extacted fom the simulated phaso voltage at nodes RF + and RF. The simulation esults ae geneated using an Agilent Advanced Design System (ADS). The C T value is set to.3 pf to obtain db amplitude eo and o phase eo. The NMOS tansisto, octagonal spial inducto, and metal-insulato-metal (MIM) capacito with paticula specifications ae used in the. μm CMOS standad pocess. V DD Load -. o, espectively. Within the MMDS band, the simulated amplitude eo is maintained at db. The simulated phase eo which deviates fom. GHz is highe than that at. GHz, but is still unde 9 o. 9 7 5 3 Amplitude Eo Phase Eo -.5.55..5.7 Fequency (GHz) Fig. 3. Simulated amplitude and phase eos between nodes RF + and RF in T netwok-based downconvete within MMDS band. db o Degee LO + RF L D M 3 M M 5 M LO - RF + RF - V REF L T C T L T M M L D V REF Fig.. Cicuit schematic of poposed downconvete with a T phase-shift netwok. 5 3 - - -3 Amplitude Eo Phase Eo IF + -...3..5. CT (pf) Fig.. Simulated amplitude and phase eos between nodes RF + and RF in T netwok-based downconvete within ange of C T values fom to. pf fo L T value of.3 nh. Fig. 3 shows the amplitude and phase eos between nodes RF + and RF in the multi-point micowave distibution system (MMDS) band that coves fequencies fom.5 to.9 GHz At the cente design fequency of. GHz, the simulated amplitude and phase eos ae db and IF - 3. Load Impedance Analysis A downconvete that is esponsible fo the fequency convesion pocess in a WiMAX eceive is commonly equied to meet the high lineaity specifications of a high dynamic ange. In designing downconvetes, designes geneally have to tade off convesion gain fo high lineaity. In fact, the tade-off between convesion gain and lineaity detemines the load at the IF output pot of the downconvete, which is indicated in Fig. by a boken-line ectangle. The impedances of two loads, an inducto and a π esonato, ae analyzed below. 3. Inducto Load Fig. shows that two inductances (L i ) is used as the load of the T netwok-based downconvete. Notice that the supply voltage V DD is egaded as altenating cuent (ac) gounded in the ac analysis. The load impedance looking into node IF + is denoted by Z L,I. Z L,I can be obtained by connecting two seies L i, and is given by Z LI, ( j Li) () whee ω is the angula fequency. L i V DD IF + IF - Z L,I Fig.. Inducto load in T netwok-based downconvete. L i
3 J.-M. WU, C.-K. LIOU, C.-J. CHUANG, Y. KUO, A. V T NETWORK-BASED. GHZ DOWNCONVERTER RFIC 3. π-resonato Load Fig. 5 shows that a π esonant cicuit that is composed of one esistance (R ), one inductance (L ), and two capacitances (C ) is used as the load of the T netwok-based downconvete. In the cicuit shown in Fig. 5, Z L,R is the load impedance looking into node IF +. The RFC denotes an RF choke. The cente node between nodes IF + and IF is vitually gounded due to the balance of the cicuit. The π esonato can be analyzed using its diffeential half-cicuit. Fig. shows the equivalent diffeential half-cicuit with an impedance of half Z L,R. Half of Z L,R is obtained by consideing the connection of R /, L /, and C in paallel: Z ( j C ) LR, R j L [ j( C )] R L. () Resonance occus when the imaginay pat of Z L,R / is zeo. The esonant angula fequency ω can be deived as. (3) L C Then, Z L,R / at ω is R /, which is a puely eal impedance. Hence Z L,R at esonance can be found as Z LR, R. () R can be used to detemine the impedance of the π- esonato load at esonance. C L R IF + IF - Z L,R V DD RFC Fig. 5. π-esonato load in T netwok-based downconvete. L / C R / IF + Z L,R / Fig.. Equivalent diffeential half-cicuit of π-esonato load. C 3.3 Compaison of Results and Discussions The IF fequency that is used in the T netwok-based downconvete is 37 MHz. Fo an inducto load, the L i in Fig. is set to 33 nh. Substituting this value into () yields a magnitude of Z L,I of appoximately.55 kω. Fo a π-esonato load, R, L, and C in Fig. 5 ae set to kω, 3 nh, and.5 pf, espectively. Substituting these values into (3) and () yields a esonant fequency of 37 MHz and a magnitude of Z L,R at esonance of kω. Notably, L of the π-esonato load is much less than L i of the inducto load, which, in this case, is /. Fig. 7 compaes the magnitudes of Z L,I and Z L,R vesus fequency. As shown in Fig. 7, the magnitude of Z L,I inceases with fequency and its value is.55 kω at an IF fequency of 37 MHz. Fig. 7 also demonstates that the magnitude of Z L,R as a function of fequency is like a spike. The maximum Z L,R magnitude is kω, which is eached at a esonant fequency of 37 MHz, which is the IF fequency. A compaison of Z L,I and Z L,R magnitudes at 37 MHz shows that the magnitude of Z L,R is much highe than that of Z L,I, by, in this instance, a facto of almost 3. This esult implies that a small inductance can yield a lage load impedance when a π esonant cicuit is used. The convesion gain fo a downconvete is usually popotional to its load impedance. The π-esonato load with high load impedance can then ensue that the convesion gain of the downconvete is high. In contast, an inducto load with a low load impedance yields a low convesion gain of the downconvete. k --- ZL,I ZL,R ýzl,i ý ýzl,r ý...3..5..7. Fequency (GHz) Fig. 7. Compaison of simulated magnitudes of Z L,I and Z L,R. Fig. plots the dynamic load line supeimposed on the dain-souce cuent (I DS ) vesus the dain-souce voltage (V DS ) cuves of tansistos M 3 and M 5 that ae displayed in Fig.. The input RF powe is set to - dbm, which is less than the input-efeed db compession point (IP db ) of the T netwok-based downconvete using an inducto load o a π-esonato load. Fig. shows that the locus of the dynamic load line fo an inducto load is an ellipse, wheeas that of dynamic load line fo a π-esonato load is a distoted ellipse. The main eason fo the distotion is that high convesion gain that is povided by the π-esonato load easily causes satuation owing to gain compession. The method given by [] fo measuing the lineaity pe-
CG (db) IDS (ma) RADIOENGINEERING, VOL., NO., APRIL 33 fomance is to estimate the locus of a dynamic load line. An elliptic locus coesponds to good lineaity. Theefoe, the T netwok-based downconvete with an inducto load can povide highe lineaity than that with a π-esonato load......... Inducto Load Resonato Load..7..9.7.7.7.73.7.75.7 VDS (V) VGS =. V VGS =. V VGS =.5 V VGS =.55 V Fig.. Compaison between the simulated dynamic load lines supeimposed on I DS -V DS cuves fo inducto and π-esonato loads. Fig. 9 compaes the simulated esults fo the gain compession of the T netwok-based downconvete with inducto and π-esonato loads. The downconvete with an inducto load yields a convesion gain of 5 db and an IP db of - dbm. The convesion gain and IP db ae 7.7 db and -5 dbm, espectively, fo the downconvete using a π-esonato load. Clealy, a π-esonato load with the lage load impedance povides a.7 db highe convesion gain fo a T netwok-based downconvete but at the cost of a 5 db lowe lineaity than that of an inducto load with the smalle load impedance. 9 7 5 3 Inducto Load Resonato Load -3-5 - -5 - -5 Input RF Powe (dbm) Fig. 9. Compaison of simulated gain compession in T netwok-based downconvete with inducto and π-esonato loads.. Implementation and Expeiment The lineaity of a downconvete is essential in WiMAX applications. Theefoe, an inducto load is used as an output inteface netwok in the implemented T netwok-based downconvete. Notably, the 33 nh inducto is aanged off-chip to educe the chip aea. The micophotogaph in Fig. shows that the T netwok-based downconvete RFIC that is implemented using a TSMC. μm CMOS foundy pocess has an aea of.93 mm.93 mm. The downconvete RFIC was mounted and bounded on an FR pinted cicuit boad fo testing. The downconvete equies a thee-pot measuement setup to povide the RF signal and LO caie with powe and fequency sweeps as input signals, and to geneate the output IF signal to evaluate all RF paametes. The single-ended RF input impedance is matched to 5 Ω by an extenal matching netwok. A micostip ing hybid is used as a passive balun to convet an incoming singleended LO signal into a diffeential one with the input impedance matched to 5 Ω. Similaly, an output IF diffeential signal is tansfomed into a signal-ended one with the output impedance matched to 5 Ω using a Mini-Cicuits TC--3M+ passive balun. The powe consumption (P DC ) fom the single supply voltage of. V is 5.7 mw. The applied LO powe is -3 dbm. The measued RF, LO, and IF etun losses exceed db, db, and db, espectively. LO + RF IF + IF - LO - Fig.. Chip micogaph of the implemented downconvete RFIC. Fig. plots the measued convesion gain and noise figue in the WiMAX MMDS band that coves fequencies fom.5 to.9 GHz. Notice that the measued convesion gain is obtained using an R&S ZVA vecto netwok analyze and an R&S SMF A signal geneato. The noise figue is measued with a Mini-Cicuits NC3C noise souce, an R&S SMBVA vecto signal geneato, and an R&S FSV signal analyze. The maximum convesion gain is 5. db at.5 GHz. As indicated in Fig., the 3 db bandwidth coves the entie applied fequency ange. Although the input stage of the T netwok-based downconvete is a passive netwok, by a multiplication stage that compises fou tansistos can povide its convesion gain. Fig. also shows that the measued noise figue is highe than db due to paasitic loss of the.3 nh octagonal spial inducto in the T netwok, but still unde 7. db fom.5 to.9 GHz. Fig. plots the measued the LO-to-IF, LO-to-RF, and RF-to-IF isolations of the T netwok-based downconvete. Within the LO fequency fom. to.3 GHz, both LO-to-IF and LO-to-RF isolations ae geate than 7 db, but below db. This shotcoming can be ovecome by keeping the incoming LO powe to the downcon-
db db 3 J.-M. WU, C.-K. LIOU, C.-J. CHUANG, Y. KUO, A. V T NETWORK-BASED. GHZ DOWNCONVERTER RFIC vete low. The RF-to-IF isolation exceeds db in the RF fequency ange that coves fom.5 to.9 GHz. CG NF.5.55..5.7 Fequency (GHz) Fig.. Measued convesion gain and noise figue in T netwok-based downconvete. - -7 - -9 - - LO-to-IF Isolation LO-to-RF Isolation RF-to-IF Isolation -...3..5..7 Fequency (GHz) Fig.. Measued isolation in T netwok-based downconvete. In a two-tone test, the RF, LO, and IF fequencies in the downconvete ae set to,, and 37 MHz, espectively. The fequency spacing is set to MHz, which is the channel bandwidth in WiMAX systems. Fig. 3 shows the simulated and measued output powes of fundamental and thid-ode intemodulation (IM 3 ) poducts in the T netwok-based downconvete. The ageement is quite good. The convesion gain and the IP db of the downconvete ae 5 db and - dbm, espectively. Fig. 3 also indicates that the T netwok-based downconvete has an input-efeed thid-ode intecept point (IIP 3 ) of dbm, evealing that it has good lineaity. This is because a passive T-type LC netwok that is used as a single-ended-to-diffeential convete eplaces the active diffeential-pai tansconductance stage of a Gilbet mixe, inceasing the voltage dop and educing the voltage swing. Tab. summaizes the pefomance meits of this wok and compaes it with pevious poposed designs of low supply-voltage downconvetes in [], [3], [5], [], [9]. The applied LO powe level that is geneated by a fequency synthesize must be consideed to detemine the powe consumption in an oveall WiMAX system, especially in the low-voltage downconvesion mixe design. In Tab., a figue of meit (FOM) in tems of LO powe is defined as CG IIP3 FOM log( ) ( ). (5) NF P P W Numeical values of CG and NF ae substituted into (5). The IIP 3, P DC, and P LO ae expessed in watt (W). The compaison shows that the poposed downconvete has the highest FOM among all the woks listed in Tab.. Output Powe (dbm) - - -3 - -5 - -7 Fundamental IM3 - -3-5 - -5 - -5 5 Input RF Powe (dbm) DC IPdB=-dBm LO Simulation Measuement IIP3=dBm Fig. 3. Simulated and measued output powes fo the fundamental and IM 3 poducts in T netwok-based downconvete. [] [3] [5] a [] [9] This Wok RF Feq. 5.5 5.5.9..5. (GHz) Supply.9.9..77.. Voltage (V) P DC (mw).57.95... 5.7 P LO (dbm) -3. 5.5 5 - -3 CG (db).9.3 3 5.7 5. 5 NF (db).5 5.. IIP 3 (dbm) -.555.3 - -5.7 -. FOM (/W) -..3 5.9 3.9.7 5.5 a: Simulated esults Tab.. Compaison of pefomance of the poposed downconvete with those of downconvetes pesented elsewhee. 5. Conclusions A low-voltage. GHz downconvete RFIC design with a T phase-shift netwok and an inductive load was implemented using a. μm CMOS standad pocess. Its main advantage is its lowe input LO powe and highe IIP 3 than those of othe available techniques. Acknowledgements This eseach was suppoted in pat by the R.O.C. (Taiwan) National Science Council unde Gant -- E-7- and by the R.O.C. (Taiwan) Depatment of Industial Technology unde Gant 99-EC-7-A--S-. The authos wish to thank the R.O.C. (Taiwan) National Chip Implementation Cente fo poviding the TSMC. μm CMOS foundy sevice.
RADIOENGINEERING, VOL., NO., APRIL 35 Refeences [] DEBONO, C. J., MALOBERTI, F., MICALLER, J. A 9 MHz,.9 V low-powe CMOS downconvesion mixe. In Poceedings of the IEEE Confeence on Custom Integated Cicuits. San Diego (U.S.),, p. 57-53. [] HUANG, M.-F., LEE, S.-Y., KUO, C. J. A 5.5 GHz even hamonic mixe fo low voltage diect convesion eceives. In Poceedings of the 5 Asian Solid-State Cicuits Confeence. Hsinchu (Taiwan), 5, p. 3-3. [3] HUANG, M.-F., KUO, C. J., LEE, S.-Y. A 5.5-GHz CMOS folded-cascode even-hamonic mixe fo low-voltage applications. IEEE Tansactions on Micowave Theoy and Techniques,, vol. 5, no., p. - 9. [] DEGUCHI, J., MIYASHITA, D., HAMADA, M. A.V 3µW dbm LO-input.GHz double-balanced cuent-eusing single-gate CMOS mixe with cyclic passive combine. In Digest of the 9 IEEE Intenational Solid-State Cicuits Confeence. San Fancisco (U.S.), 9, p. - 5,5a. [5] JAFFERALI, N., DEEN, M. J. Low-voltage and low-powe.9ghz body-input downconvesion mixe. In Poc. of the Canadian Confeence on Electical and Compute Engineeing. Niagaa Falls (Canada),, p. 3 -. [] LIANG, K.-H., CHANG, H.-Y., CHAN, Y.-J. A.5 7.5 GHz ulta low-voltage low-powe mixe using bulk-injection method by.-μm CMOS technology. IEEE Micowave and Wieless Components Lettes, 7, vol. 7, no. 7, p. 53-533. [7] LIANG, K.-H., CHANG, H.-Y..5- GHz low-voltage low-powe mixe using a modified cascode topology in. μm CMOS technology. IET Micowaves, Antennas & Popagation,, vol. 5, no., p. 7-7. [] HERMANN, C., TIEBOUT, M., KLAR, H. A.V.mW tansfome based.5ghz downconvesion mixe with +5.dB gain and -.dbm IIP3 in.3μm CMOS. In Digest of the IEEE Radio Fequency Integated Cicuits Symposium. Fot Woth (U.S.),, p. 35-3. [9] HERMANN, C., TIEBOUT, M., KLAR, H. A.-V.-mW tansfome-based.5-ghz downconvesion mixe with +5.-dB gain and -.-dbm IIP3 in.3-μm CMOS. IEEE Tans. on Micowave Theoy and Techniques, 5, vol. 53, no., p. - 95. [] YANG, T.-Y., TU, H.-L., CHIOU, H.-K. Low-voltage high-linea and isolation tansfome based mixe fo diect convesion eceive. In Poc. of the IEEE Intenational Symp. on Cicuits and Systems. Island of Kos (Geece),, p. 375-3757. [] LAI, J.-T., LIN, Y.-S., LU, C.-L., CHUANG, H.-R. A 3-5-GHz low-voltage high-isolation tansfome-based CMOS mixe fo UWB applications. In Poc. of the 3 d Int. Conf. on Innovative Computing Infomation and Contol. Dalian (China),, p. 3. [] OHARA, S., NAKASHA, Y., IWAI, T., JOSHIN, K. Measuement of dynamic load lines of powe heteojunction bipola tansisto. In Digest of the 997 IEEE MTT-S Intenational Micowave Symposium. Denve (U.S.), 997, p. 99-9. About Authos... Jian-Ming WU was bon on Novembe 3, 97, in Kaohsiung, Taiwan. He eceived the B.S.E.E degee fom the Yuan Ze Univesity, Chungli, Taiwan, in 997, and the M.S.E.E. and Ph.D. degees fom the National Sun Yat- Sen Univesity, Kaohsiung, Taiwan, in and, espectively. He joined the Gaduate Institute of Communication Engineeing, National Univesity of Tainan, Tainan, Taiwan, as an assistant pofesso in. In 7, he joined the Depatment of Electonic Engineeing, National Kaohsiung Nomal Univesity, Kaohsiung, Taiwan, as an assistant pofesso. His cuent eseach inteests include design of monolithic micowave integated cicuit, modeling of RF components, and chippackage-boad codesign. He was the ecipient of the National Sun Yat-Sen Univesity Outstanding Doctoal Dissetation Awad. Since, he has been seving on the IC Implementation Review Committee, National Chip Implementation Cente, Hsinchu, Taiwan. Ching-Kuo LIOU was bon on Decembe 3, 95, in Penghu, Taiwan. He eceived B.S.E.E. degee fom the Depatment of Electonics, Cheng Shiu Univesity, Kaohsiung, Taiwan, in, and the M.S.E.E. degee fom the Depatment of Electonic Engineeing, National Kaohsiung Nomal Univesity, Kaohsiung, Taiwan, in. His eseach is focused on the design of low-voltage downconvesion mixes. Ching-Jui CHUANG was bon on Octobe, 95, in Taoyuan, Taiwan. He eceived B.S.E.E. degee fom the National Taiwan Ocean Univesity, Keelung, Taiwan, in, and is cuently woking towad the M.S. degee in electonic engineeing, at the National Kaohsiung Nomal Univesity, Kaohsiung, Taiwan. Yujin KUO was bon on July, 3, 93, in Keelung, Taiwan. He eceived B.S.E.E. degee fom the National Kaohsiung Univesity of Applied Sciences, Kaohsiung, Taiwan, in, and the M.S.E.E. degee fom the Gaduate Institute of Communication Engineeing, National Univesity of Tainan, Tainan, Taiwan, in. He is cuently a Senio Enginee with the R&D Cente, Communication and Visual Poducts, HW RF Sec., Pegaton & Unihan Copoation, Taipei, Taiwan. His eseach is focused on the design and testing of RF components fo DVB, ATSC, and WiFi applications.