' $ Predistortion and Postdistortion Linearization of RF Ampliers Patrick Roblin Dominique Chaillot Sukkuen Myoung Xian Cui Mamta Debnath March 18, 2004 Department of Electrical and Computer Engineering & Microwave Laboratory The Ohio State University Columbus, OH 43210 T . H . E OHIO STATE UNIVERSITY The Ohio State University % ' $ Introduction: Design Trends Pull for more power ecient ampliers { Class B,C, D and E ampliers are more ecient than class A but exhibit poorer linearity { call for improved linearization performance Push for lower cost linearization systems { Adaptive predistortion linearization replacing feedforward { Alternative novel baseband linearization possible Increasing signal bandwidth (e.g. from 2.5G to 3G) { Need to deal with both slow (thermal drift) and fast (elecrical) memory eects { Electrical memory eects degrade the ACPR for wide-bandwidth signals & Technique Correction Analog Predistortion 3-5 dB Cross Modulation 15-20 dB Feed Forward 30 dB Digital Predistortion 15-20 dB Microwave Laboratory Bandwidth 15-25 MHz 10-20 MHz 25-60 MHz 15-20 MHz Eciency 5-8 % 10-12 % 6-10 % 12-14 % Flexibility Cost Low Low Medium Medium Medium High High Medium T . H . E OHIO STATE UNIVERSITY The Ohio State University % ' $ Memory Eects and Volterra Modeling A memoryless SISO system can be modeled by: y(t) = gm x(t) + gm3x3(t) The most rigorous theory for including memory eects is the Volterra formalism. In that formalism the system is described by Volterra kernels of various orders. For example a SISO system with a 3rd order kernel is of the form: ZZZ 1 y(t) = h(1; 2; 3)x(t ; 1)x(t ; 2)x(t ; 3)d1d2d3 ;1 & Microwave Laboratory T . H . E OHIO STATE UNIVERSITY The Ohio State University % x æ á{ ~ æ } x z x z y z è{ ç â B F 8 D > 6 =6 ? :? = = ê ý 6 ; 6 :;8 6ë 6 :; =C ? > ë < C >; :; 6 C;8 >; DF > 6 :; 6 ; ë 6 8 D 8 D> :D : : C= = ëê <6 üý C= < A8 = C 6 B 6 = >; > ; 8 6 ë :?> = >7 6ë :ê B F 6 =6 ? ; :6 ý 6 ë C:D = >E ð6 :8 CýB ë ý < 6 D :; = ï8 =C ö " Ó ! ! #$Ó Ô Û õ ò Ô # ö % Ó Ó ! ! Û õ ò 7 Aê 6 C ; 6B 6ë ) : 5 í* 8 D (ì = (ì ; '6 ý :?> ;ë C=6 B B> ; 6 &6 ý 6 ðäê ï =6 C 6ë 6 A ë ; <6 ý 6 :8 é F7 F ; 8 ; 6 D > 6ë ; 6> ; ;> ; +6 ý ? ý = C :6ë D ; 6 > 6 üý 6 8Û D :8 Û % = " : ý G # , , , û Ó Ó Ó ß Ó Óû # Ó Ô ö ! Ð Û õ ò # Ó Ó ß Ó Ó Óû ý < 6 ; ?> 6 :D 6 C 8 ï 6 :D = B ýCB ý ;ë B C=6 . = B> 8 6 ß ß Û 10Û , , D :8 . &6 û 0 . . . Û 6 ö #/Ó Ô Ð õ ! ò A = :2 6 6 8 :6 6 :A; >ÿ : ý T . H . E OHIO STATE UNIVERSITY p d n fvk d ujt k ins n d gqr k q f ing f g im defg hij lk c o o ' $ Memory Eects A PA system is memoryless if all coecients are real, e.g. (LSB & USB IMD): A0 = B0 = 34 gm3 A PA system is quasi-memoryless if the coecients are complex and frequency independent, e.g. (LSB & USB IMD): A0 = B0 = 34 ym3 = jym3j exp(jm) A PA system exhibits memory eects if its coecients are frequency dependent, e.g. (LSB & USB IMD): A0 6= B0 or ym3(;) 6= ym3 (+) & Microwave Laboratory T . H . E OHIO STATE UNIVERSITY The Ohio State University % ACPR Degradation Up to 17 dB degradation of ACPR at low and high bandwidth 27 ' $ RF predistortion (quasi-memoryless algorithm) VDC RF(I,Q) RF IQ Modulator RF(I’,Q’) α Baseband 2 VE 0110 1010 1010 10 RF PA β 1 2 1 2 α=1+ α3 VE +α 5 VE β=1+ β 3VE +β5 VE Polynomial functions & Microwave Laboratory T . H . E OHIO STATE UNIVERSITY The Ohio State University % ' $ Adaptive baseband digital predistortion (quasi-memoryless algorithm) I DAC Q DAC α RF β LO 2 2 I +Q LUT Adaptation Algorithm ADC RF ADC 2 The adaptation can only take care of slow memory eects The linearization algorithm cannot handle fast memory eects & Microwave Laboratory T . H . E OHIO STATE UNIVERSITY The Ohio State University % ' $ RF Predistortion for Quasi-Memory less Systems For Two-tone Excitations: Volterra analysis assumption : A4 = B4 = 0 (5th order noise) 3 and 3 are real coecients for Quasi-memoryless systems " 3 = ;3Re (1 + j ) yym3 m & Microwave Laboratory # and " 3 = ;3Im (1 + j ) yym3 m # T . H . E OHIO STATE UNIVERSITY The Ohio State University % z{ á à Zy m y á 7 â â z{ 4 x{ y z{ y ð | z è`æ ë >; B C8; = C ~ z{ y ç @6 = > >; ü :> ö ö ^ 6 Cë => 6 :D ; :> ý % " ð ? 8 õ dcò b ß ß :> Cë ë Aê = 8 8 :6 A:; >ÿ ^ = í C? D 8 D 6 ; g f 7 ýë 6 ë 8 ý7 ;ë C=6 B &6 B> A6 < B> 6 D :8 Ð = 8Ð e ^ < ? :? D 6> 6 > 6ë 6 D ; 6 :6 ý ï F D> A6ë 5 >; ÿ +ÿ = C :>; 6 D Cë :A C8 :>; 6B 7 G õ ò k k n n Ó Ô Óû Ô ö Ð n n ö Ð e f õ ò õ ò O ýC ; % # " Û Û k û ö Ô n k ö Ó n O O õ ò õ ò # . , Û Û T . H . E OHIO STATE UNIVERSITY p d n fvk d ujt k ins n d gqr k q f ing f g im defg hij lk c o o ' $ Two Band Implementation The I and Q signals need to be phase shifted to account for memory eects: RF(I,Q) RF IQ Modulator RF(I’,Q’) RF PA α β 1 2 Baseband 2 & Microwave Laboratory VE 1 2 1 2 α=1+ α3 VE +α 5 VE β=1+ β 3VE +β5 VE Polynomial functions T . H . E OHIO STATE UNIVERSITY The Ohio State University % ¢ u ¤ ¡ ´ u Amplitude (dB) ª £ ´« ¢ § u § ¡ t ¥ Amplitude (dB) x{ y æ y ¢ . 280 H . E OHIO STATE 275T 270 265 255 260 Frequency (MHz) 250 245 240 235 s § −50 ¯ ¦ ¢ ¥ t ¡ −50 r r q © 0 á ~ p xæ zá{ z4 ` y x{ y è à} ± ¢ −100 −100 −150 262 260 258 256 Frequency (MHz) 254 252 250 −150 « 0 UNIVERSITY p d n fvk d ujt k ins n d gqr k q f ing f g im defg hij lk c o o ' $ Predistortion Using PA Model PA modeled as multiple 3rd order systems in parallel: H30 H34 H31 H33 H32 x H30’ y H34’ H31’ H33’ H32’ & Microwave Laboratory T . H . E OHIO STATE UNIVERSITY The Ohio State University % ' $ Iterative Predistortion for 8 tone Multsine Model is a 3rd order Volterra system The correction can be applied iteratively to remove higher order terms generated Normal PA output Ouput for QOBM Output for Predistortion 0 Amplitude (dB) −50 −100 & Microwave Laboratory T . H . E −150 235 240 245 250 255 260 Frequency (MHz) 265 270 275 280 OHIO STATE UNIVERSITY The Ohio State University % ' $ Novel BaseBand Linearization Schemes Low frequency feedforward linearization (LFFF) Input and Output baseband modulation (IBM and OBM) Bilateral and Quadratic baseband modulation (BBM and QOBM) & Microwave Laboratory T . H . E OHIO STATE UNIVERSITY The Ohio State University % ' $ Low frequency feedforward linearization 2 2 3 Ids = gm1 vgs + gd1 vds + gm2 vgs + gm1d1 vgs vds + gd2 vds + gm3 vgs 2 2 3 4 3 +gm2d1 vgs vds + gm1d2 vgsvds + gd3 vds + gm4 vgs + gm3d1 vgs vds 2 2 3 4 +gm2d2 vgs vds + gm1d3 vgsvds + gd4 vds + ; (1) where Ids = drain-source current, vgs = gate-source voltage, vds = drain-source voltage Port 1 Port 2 PA fa fb RF IMN RF a1 b1 OMN RFC c 2 & Microwave Laboratory 2 A Baseband fc = fb − fa T . H . E OHIO STATE UNIVERSITY The Ohio State University % ' $ Input and Output Baseband Modulation Linearization of PAs New proposed baseband linearization schemes (IBM,OBM and BBM) Square law operation is performed directly on the baseband Filter no longer required Port 1 LO Port 2 RF PA Baseband RF RF IMN I&Q RF a1 b1 OMN RFC c & Microwave Laboratory IBM I 2+ Q 2 A 1 c 2 OBM T . H . E OHIO STATE UNIVERSITY The Ohio State University % ' $ Perfect IM3 and IM5 cancellation using gmd for memoryless systems If the RF signal is given by vin (t) = I (t) cos(!t) + Q(t) sin(!t) the correction for a 5th order system with gm3 and gm5 terms is 3 g 5 gm5 [I 2 (t) + Q2 (t)]2 vLFFF (t) = ; m3 [I 2 (t) + Q2 (t)] ; 4 gmd 8 gmd 0 −50 −100 Amplitude (dB) −150 −200 −250 −300 & Microwave Laboratory −350 Input Output without Feed Forward Output with Feed Forward −400 T . H . E −450 235 240 245 250 255 260 Frequency (MHz) 265 270 275 280 OHIO STATE UNIVERSITY The Ohio State University % ' $ Constraints required by Low Frequency Feedforward jGj d ; m m3 j = 23 jjyymd j = (2m + n + 1) = n (2) Phase shift m and d associated with gm3 and gmd respectively (cancellation of the 3rd order intermodulation at RF frequencies) ym3 = jym3j exp(jm ) ymd = jymd j exp(jd) & (3) (4) (5) New phase constraint d ; m = n required (Yu and Roblin, 2001) Microwave Laboratory T . H . E OHIO STATE UNIVERSITY The Ohio State University % ' $ IMD plotted versus phase for dierent gains G (S. Myoung) xxx −2 −2 −35 −4 −2 −10 −4 −8 −4 −6 −6 −10 −2 −40 −1 −4 0 −8 −10 −6 −50 −6 IMD (dBc) −8 −45 −55 −8 −60 & Microwave Laboratory 0 50 100 150 200 phase (deg) 250 300 350 T . H . E OHIO STATE UNIVERSITY The Ohio State University % ' $ Experimental Setup for Verication (Xian Cui) LO PA 0 1 2 cos wt RF IMN LO rejection mixer AWG Baseband Arbitrary wave generator RF a1 b1 OMN RFC IBM RFC c 1 c 2 OBM A cos (2wt + φ ) 012 The LO rejection mixer and AWG are controlled by a PC using Labwindow. & Microwave Laboratory T . H . E OHIO STATE UNIVERSITY The Ohio State University % ' $ Experimental Results for IBL −25 −30 −35 IMD power −40 −45 −50 −55 −60 −65 & −70 Microwave Laboratory 0 50 100 150 200 Phase 250 300 350 400 T . H . E OHIO STATE UNIVERSITY The Ohio State University % ' $ Experimental Results for OBL −35 −40 IMD power −45 −50 −55 −60 −65 & −70 Microwave Laboratory 0 50 100 150 200 Phase 250 300 350 400 T . H . E OHIO STATE UNIVERSITY The Ohio State University % ' $ Limitations of IBM, OBM (LFFF) IBM and OBM do not provide enough degrees of freedom to eliminate both LSB and USB intermodulation in two-tone exitations. New methods with more degrees of freedom will be investigated next to cancel both LSB and USB: { Bilateral Baseband modulation (BBM) { Quadratic Output Baseband modulation (QOBM) { Quadratic Output Baseband modulation (QIBM) & Microwave Laboratory T . H . E OHIO STATE UNIVERSITY The Ohio State University % ' $ Bilateral Baseband Modulation (BBM) for PA with Memory Port 1 LO PA Baseband RF IQ Port 2 IMN RF a1 b1 OMN RFC RFC c 1 c2 A1 I2+ Q2 1 A2 BBL 2 & Microwave Laboratory T . H . E OHIO STATE UNIVERSITY The Ohio State University % ' $ Volterra Analysis of BBM for Systems with Memory X1 H30 H30 a1 b1 H34 H31 H34 H31 a21 b1* a1* b12 H33 H32 a21 b1* a1* b12 H33 H32 X a1 b1 c1 H20 H20 H22 H21 X2 a1 c*1 b1 c1 H22 H21 c2 a1 c*2 b1 c2 For a two-tone excitation the intermodulation terms of interest are: iout (2!1 ; !2 ) = A0a1 2 b1 + A1a1 c1 + A2 a1c2 + A3b1 c1 2 + A4b1 c2 2 + A5 b1 c1 c2 iout (2!2 ; !1 ) = B0a1 b1 2 + B1b1 c1 + B2 b1 c2 + B3 a1c1 2 + B4 a1 c22 + B5a1 c1c2 where A0 A5 and B0 B5 are the coecients for the LSB and USB IMD3 respectively. The two desired tones at the output of the PA are themselves given by: iout (!1 ) = D0 a1 + D1b1 c1 + D2b1 c2 + D3a1 2 a1 + D4a1 b1 b1 + D5 c1c1 a1 +D6 c1 c2a1 + D7 c1c2 a1 + D8 c2 c2 a1 iout (!2 ) = E0 b1 + E1 a1 c1 + E2 a1 c2 + E3 b1 2 b1 + E4 b1 a1a1 + E5 c1 c1 b1 T . H . E OHIO +E6 c1 c2 b1 + E7 c1c2 b1 + E8 c2 c2 b1 & Microwave Laboratory STATE UNIVERSITY The Ohio State University % ' $ Volterra Analysis of BBM for Systems with Memory Power Swept Response Power Swept Response −50 −50 −60 −100 −70 −150 Harmonic current (dB) Harmonic current (dB) −80 −90 −100 −110 −120 −200 −250 −300 −130 −350 −140 −150 −15 & −10 −5 0 5 Power (dBm) 10 15 20 −400 −15 −10 −5 0 5 Power (dBm) 10 15 20 Under certain operating conditions both the upper and lower IMD3s can be theoretically reduced by 100dB and 200 dB respectively. Note however that, depending on the Volterra Kernel, the biquadratic system does not always admit a solution for certain power levels. Microwave Laboratory T . H . E OHIO STATE UNIVERSITY The Ohio State University % z{ á à Zy m á â ù øøù IMN y 7 ý üüý OMN a1 b1 RF â z{ 4 à o¡ RFC PA 90 cQ AQ Q Baseband a ÷ öö÷ û úúû 2 2 I+Q RF I&Q 4{ â æ èæ x 6 z z y è{ ç IMN cI AI I OMN a1 b1 RF LO PA RFC ï ÿ u ¤ ð t þ Ó § § Ð f Ó # Ó D ö = 8 Ð 7 f Ó # Ó e ö < 8Ð 6 6 >; D 6 e D ?> :Ð = = = C :>; 6 D Cë : 6 :?> C ' f äðê ï =6 C 6 D :8 Ð = 8Ð e <6 ; ê ëê ë ë A6ë :> <6 <C ë 8 F 8 :> G T . H . E Ð ùñ Ð ùñ k < í D 8 û û ö Ð = k 6 O O õ ò õ ò UNIVERSITY ö Ð e h f OHIO STATE ùñ ùñ l j l j p d n fvk d ujt k ins n d gqr k q f ing f g im defg hij lk c o o ' $ Multitone Performance Limitation Consider a 2 band case (dierent phase for LSB and USB) with 2-tone and multitone excitations. Works for 1 tones per frequency band but degrades for multitones Disadvantage: the number of linearization constraints increases with the number of frequency bands Advantage: no PA modeling required 0 0 −50 −50 −100 Amplitude (dB) Amplitude (dB) −100 −150 −200 & −150 −200 −250 −250 −300 −300 −350 250 252 Microwave Laboratory 254 256 Frequency (MHz) 258 260 262 −350 T . H . E 940 960 980 1000 1020 1040 Frequency (MHz) 1060 1080 OHIO STATE 1100 1120 UNIVERSITY The Ohio State University % ' $ QOBM Using a PA Model Model is a 3rd order Volterra system Higher order term in the baseband output modulation are neglected Normal PA output Ouput for QOBM Output for Predistortion 0 Amplitude (dB) −50 −100 & Microwave Laboratory T . H . E −150 940 960 980 1000 1020 1040 Frequency (MHz) 1060 1080 1100 1120 OHIO STATE UNIVERSITY The Ohio State University % ' $ Testbed for Digital Vectorial Predistortion, BBM and QOBM PC Labwindow PC control Adjustable Gain and Offset I Programation 12 bits ADC I RFout Diff. OpAmp Diff. OpAmp FPGA Diff. OpAmp Q I DAC LO Q Adjustable Gain and Offset 14 bits I LO/RF in Diff. OpAmp 14 bits 12 bits RFout Q Q DAC ADC Adjustable Gain and Offset Adjustable Gain and Offset Clock Distribution 80 MHz Clock Goals: Study impact of memory eects for broadband signals Compare the performance of the various linearization schemes Study potential combination of linearization techniques & Microwave Laboratory T . H . E OHIO STATE UNIVERSITY The Ohio State University % ' $ Capabilities of the LSNA 70 dBc dynamic range -80 dBm noise oor Calibrated in frequency, time and power Measure amplitude and phase of up to 10,000 tones Acquire harmonics up to 20 GHz (in OSU system) 4 channels, 10 W damage level Equipped with a load pull Modulated source (E4438C-504) 250 kH to 4 GHz (with WCDMA 3GPP) & Microwave Laboratory T . H . E OHIO STATE UNIVERSITY The Ohio State University % ' $ Applications of the LSNA in PA Design and Linearization PA design: { Model verication { Design PA using load pull { Tune PA (IF and harmonics impedance terminations) by directly inspecting the load-line PA Modeling and Linearization: { Extract a Volterra Kernels coecients using multitone exitations { Acquire the trajectory of I and Q at the PA output directly from the RF & signal Microwave Laboratory T . H . E OHIO STATE UNIVERSITY The Ohio State University % ' $ LSNA Testbed DUT LSNA RF Source Coupler Coupler RF b1 10 MHz Synchronization a2 b2 step attenuators SRD fSRD a1 comb FracN 10 MHz fIF 20 MHz & Microwave Laboratory Clock ADCs (Spectrum MI4022) PC T . H . E OHIO STATE UNIVERSITY The Ohio State University %