Linearization Techniques for Power Amplifiers at the Device and Circuit Level (invited)

Transcription

1 Linearization Techniques for Power Amplifiers at the Device and Circuit Level (invited) Leo de Vreede PA Workshop, San Diego 2005 January 30, DIMES

2 Introduction Improving for the linearity/efficiency trade-off is complex! (it involves system, circuit & process technology considerations;) Required: Deep knowledge of device non-linearities + Profound knowledge of RF circuit design + Non-linear RF characterization tools & models Improved Circuit & Technology Design for Linearity January 30,

3 Introduction: Linearization Basics y=a 1 x (linear response) x y T 1, T 2 : desired tones P out [dbm] D f T 1 T 2 January 30, f

4 Introduction: Linearization Basics y=ax + a 2 x 2 (2 nd -order intermod.) x y T 1, T 2 : desired tones IM 2 : 2 nd -order intermod. HT 2 : 2 nd -order intermod. P out [dbm] Base band D f 2 nd harm. band IM 2 HT 2 IM 2 HT 2 T1 T2 January 30, D f f

5 Introduction: Linearization Basics y=ax + a 2 x 2 + a 3 x 3 (3 rd -order intermod.) P out [dbm] x y D f direct mixing T 1, T 2 : desired tones IM 2 : 2 nd -order intermod. HT 2 : 2 nd -order intermod. IM 3 : 3 rd -order intermod. IM 2 IM 3 IM 3 HT 2 T 1 T 2 HT 2 IM 2 January 30, D f f

6 Introduction: Linearization Basics y=ax 1 + a 2 x 2 + a 3 x 3 Indirect mixing x y January 30,

7 Introduction: Linearization Basics y=ax 1 + a 2 x 2 + a 3 x Higher order terms (x) Low power High power P out [dbm] x y D f T 1, T 2 : desired tones IM 2 : 2 nd -order intermod. HT 2 : 2 nd -order intermod. IM 3 : 3 rd -order intermod. IM 5 IM 5 IM 2 IM 3 IM 3 HT 2 HT 2 T 1 T IM 2 2 January 30, D f f

8 Introduction: Linearization Basics IM 3 cancellation is achieved when direct and indirect products have equal amplitude and opposite phase, this can be controlled by the out-of-band impedances. P out [dbm] x y D f HT 2 HT 2 IM 2 IM 3 IM 3 Base-band T 1 T 2 2 nd -harmonic impedance impedance January 30, When out-of-band terminations are properly set f

9 Overview Introduction Characterization Predictive Modeling & Device Optimization BJT based PA design FET based PA design Conclusions Outlook: Adaptive PA s Acknowledgements January 30,

10 Characterization Large Signal Device Characterization setups at the TUDELFT (custom-built) Isothermal measurements (Data base model extraction / model verification) Differential load-pull measurements (testing differential PA s) On-Wafer Active harmonic load-pull measurements (Linearity optimization using out-of-band terminations) (Evaluation Linearity new process generations) January 30,

11 Characterization: Active Load-Pull setup, Principle Large modulation bandwidth Wave form reconstruction Active load & source pull at f BB, f 0 and f 2nd INPUT MATCHING At: f 0,f 2nd and f BB c o OUTPUT MATCHING At: f 0,f 2nd and f BB High dynamic range Isothermal measurements Real-time calibration at all frequencies January 30, Novel setup for on-wafer non-linear device characterization using optimum circuit conditions

12 Characterization: Active Harmonic Load-Pull System On wafer large-signal device characterization!!! January 30, Thesis work M. Spirito + support by Marco Pelk Happy supporters from Industry

13 Characterization: Active Harmonic Load-Pull System Pload [dbm] IF bias-t 2 nd IF optimum 2 nd IF and 2 nd harm. optimum Tones Im3 see Fig.6 harm KHz W harm 50 W 1.29MHz Next channel Reduced Channel-tochannel Interference By improved linearity -70 1,9580G 1,9590G 1,9600G 1,9610G 1,9620G freq [Hz] Linearity imp. using optimum 2 nd and IF impedances for an IS-95 signal. On-wafer testing of device linearity using: January 30, optimum harmonic terminations - wideband modulated signals

14 Overview Introduction Characterization Predictive Modeling & Device Optimization BJT based PA design FET based PA design Conclusions Outlook: Adaptive PA s Acknowledgements January 30,

15 Predictive Modeling & Device Optimization Compact models Mixed level models Database models Advantages + Speed + Accuracy + Physically based + Suitable for device innovation + Physically based + Device innovation + Meas./sim. based + Speed + Accuracy + very physical Disadvantages - Difficult to extract +/- Accuracy - Speed - No device innovation - Not meas. based - Speed Vittorio Cuoco January 30,

16 Smoothie model principles: Currents & charges as line integral of the y-parameters over bias: V ds Predictive Modeling & Device Optimization Integration path V gs 1.6E-04 Drain current [A] 1.4E E E E E E E E Drain bias [V] üconsistency between DC, AC and HB simulations! üincreased order of continuity But what about rotation and thermal effects? January 30,

17 Predictive Modeling & Device Optimization The database model Smoothie Smoothie schematic Large signal equivalent circuit Smoothie, including: dynamic thermal effects improved freq. dependency Blanked to avoid prepublication à Solves for the rotation problem! à No implicit assumptions on the device behavior! January 30,

18 Predictive Modeling & Device Optimization The database model Smoothie Approx. Meas. 21 Im(Y 5,5E-05 Taylor coefficients Taylor coefficients ØImmunity to noise ØSmooth derivatives up to 5th order ØSparse grid ) Advantages of smoothing splines: Id g1 g2 g3 g4 g5 3,5E-05 1,5E-05-5,0E-06-2,5E-05 0 January 30, Gate bias 2 [V] 3 Gate bias [V] 4 Gate bias [V] ØReduced extraction time ØCompact model databases 18

19 Predictive Modeling & Device Optimization Smoothie used for device engineering IM3 [dbc] ( Philips Gen 4 LDMOS) Definition Device Structure S i m u l a t i - M e 1 a s u r 5 e m Device simulator (MEDICI) Layout extractor (Momentum) Thermal network (FEMLAB) Smoothie Data base January 30, 2006 P 19 o [ d u B Predicted and measured LDMOS Linearity! Circuit simulator (ADS) Thermal Network Layout parasitics

20 Predictive Modeling & Device Optimization Smoothie results (Figures on courtesy of V. Cuoco and P. Hammes.) GEN4 GEN5 Comparison of predicted (Smoothie) and measured IM3 for the fabricated Philips LDMOS devices GEN4 and GEN5 January 30,

21 Predictive Modeling & Device Optimization LDMOS Device innovation (Figure on courtesy of V. Cuoco and P. Hammes.) Improved linearity/efficiency trade-off Linearity versus efficiency improvements for the Philips LDMOS generations (1-2W devices, on-wafer load-pull measurements). January 30,

22 Overview Introduction Characterization Predictive Modeling & Device Optimization BJT based PA design FET based PA design Conclusions Outlook: Adaptive PA s Acknowledgements January 30,

23 BJT Based class-ab PA Design Linearity optimization (theory) Applying a out-of-band short Z L,BB =Z L,2nd =0 at DUT output simplifies analysis and provides Z S,BB and Z S,2nd to enforce perfect IM3 cancellation [1]: DUT Source Load Z S, B B S, 2nd C = 2tg = C I c, qpot = Z = 2t 2 b 2g je F md E = V T C j E F m January 30, F à Real part cancellation condition à Imaginary part cancellation condition à Fixed by technology [1] M.P. van der Heijden et al. BCTM 2004

24 BJT Based class-ab PA Design Low-power Linearity opt. (measurement) By sweeping Z S,BB =Z S,2nd we can experimentally find the symmetrical IM3 cancellation condition, and the related optimal quiescent current I cq. OIP3up, OIP3lo [dbm] OIP3 lower band OIP3 upper band Z S,BB =Z S,2nd [Ohm] January 30, M. Spirito,BCTM 2005

25 BJT Based class-ab PA Design IM3 vs. PAE optimization [dbc] ZL,BB and ZL,2nd (short condition) Increasing ZS,2nd Decreasing Ic Increasing ZS,BB Icq ZL,BB ZL,2nd ZS,BB ZS,2nd Case 1 Case 2 Case3 12 ma 7 ma 6.2 ma ZS,BB January 30, W 26 W 29 W ZS,2nd 17.5 W j j -25 Reference case -30 (Classical class-ab) complex ZS,2nd Purely Ohmic -70 ZS,BB=ZS,2nd=17.5 Ohm PAE [%] M. Spirito,BCTM

26 Overview Introduction Characterization Predictive Modeling & Device Optimization BJT based PA design FET based PA design Conclusions Outlook: Adaptive PA s Acknowledgements January 30,

27 FET Based Design: Lowering P back-off while still meeting the linearity spec. Derivative superposition method (most effective in back-off region) M.v.d. Heijden, IMS2001 Derivative superposition + 2 nd harmonic input tuning (Effective up to close to compression) E.W. Neo, EuMC nd harmonic input and output tuning (Effective up to the compression region) D. Hartskeerl, RFIC2005 Increased PAE January 30,

28 FET Based Design: A Square-Law Optimized Class-AB LDMOS Power Amplifier Pre-match fundamental S21 [db], S21 [deg] Before optimization IM IM3 [dbc] Active devices Harmonic f IF and f 2nd LDMOS amplifier with trans-conductance shaping using VG-offsets January 30, Pout [dbm] (M. P. Heijden, Honorable mention at MTT-S2001) Pout [dbm] After optimization 20dB imp. IM S21 [db], S21 [deg]

29 FET Based Design: A Square-Law Optimized Class-AB LDMOS Power Amplifier Transconductance Optimization Linearity optimization by controlling the higher order derivatives of the active device, through: V gs offsets (biasing) V T shifts (build-in) g m [A/V], g m3 [A/V 3 ], g m5 [A/V 5 ] ò I ( V ) gq= ud av d r a t i c D S G S m G S g m g g m5 m3 V GS [V] NEW LDMOS generation with build-in optimized Transconductance, introduced at MTT-S 2003 January 30,

30 FET Based Design DS & 2 nd Harmonic Input Tuning # IS-95 CDMA signal on Philips GEN 3 LDMOS P1dB 4 B.O Able to drive PA 2dBm harder for this linearity spec Better linearity achieved in B.O January 30, # Edmund Neo, EuMC2004

31 FET Based PA Design 2 nd Harmonic input & output tuning Measured PAE linearity trade-off for 2mm LDMOS device using: Open-Short (squares) Open-Open (circles) Short-Short (triangles) Short-Open (diamonds) 2nd harm. terminations at the device in- and output (Dave Hartskeerl, RFIC 2005) IMD3 (dbc) O-O +50% January 30, S-S O-S S-O PA E (%)

32 FET Based PA Design 2nd Harmonic input & output tuning ACPR (dbc) Measured using the TUDELFT Active Harmonic Load Pull System PAE (%) Single carrier IS-95 CDMA signal (open symbols) and single carrier 3GPP WCDMA signal (closed symbols). (D. Hartskeerl, RFIC2005). January 30,

33 Overview Introduction Characterization Predictive Modeling & Device Optimization BJT based PA design FET based PA design Conclusions Outlook: Adaptive PA s Acknowledgements January 30,

34 Conclusions A research strategy for improve device performance at the TUDelft has been presented, yielding: Unique Measurement equipment Predictive database models Linearity & PAE device optimization strategies As result Modified Class-AB device operation is still a competitive option for commercial applications providing high linearity performance at a very reasonable PAE level, while being: Low cost, Reliable, Simple... But àcurrent research fields at the TUDelft also include: Device and Circuit Innovation for Doherty PA s, Adaptive / Dynamic Load line PA s, and Polar PA s January 30,

35 Overview Introduction Characterization Predictive Modeling & Device Optimization BJT based PA design FET based PA design Conclusions Outlook: Adaptive PA s Acknowledgements January 30,

36 Outlook: Multi-mode, Multi band PA s High performance varactors for RF adaptivity January 30, Band-switching & Collector efficiency improvement with adaptive matching networks

37 Acknowledgement Philips Semiconductors: R. Jos, P. Lok, F. van Straten P. Hammes, F. v. Rijs, J. Gajadharshing. A. de Graauw, S. Theeuwen Philips Research; D. HartsKeerl, Reinout Woltjer Infineon: J.E. Muller Skyworks: P. Zampardi Many people of Agilent, BSW, Maury etc. UCSD L. Larson & P. Asbeck In special (in alphabetic order): V. Cuoco, M. v.d. Heijden, E. Neo, M. Pelk, and M. Spirito of the TUDelft. January 30,