Non-linearities characterization and modeling for system level simulations

Transcription

1 Non-linearities characterization and modeling for system level simulations Giovanni Ghione Daniel Bustos Vittorio Camarchia Simona Donati Marco Pirola Dipartimento di Elettronica Politecnico di Torino Microwave & RF electronics group

2 Outline Behavioral nonlinear models background Ongoing Activities Simulation examples: quasi memory-less models Model review Circuit level and system level simulations Comparison of various system level tools Simulation examples: models with memory (ongoing) Model description and extraction Circuit level and system level simulations

3 Device models: from physical to behavioral From: D.Root et al., IMS2004 WME-4

4 From Circuit Level to System Level Models used to simulate the PA at a component level (circuit) are capable of representing nonlinearities at a higher degree of accuracy Memory effects in principle included, although difficulties are related to model extraction (from measured data or physics-based simulations) and simulation technique used (HB vs. HB envelope) Models used to include nonlinearities at the system level are often restricted to input/output representation of (envelope) modulated baseband signals Difficulties related to identification of accurate nonlinear memory models The key issue is to link the circuit and system level models Simulation tools capable of joint circuit/system level simulations are used for validation > AWR Microwave Office + Virtual System Simulator (MWO/VSS)

5 Simulation tools At the circuit level simulation tools include Agilent ADS AWR MWO At the system level simulators used are: AWR VSS MATLAB Simulink MHOMS (C++ built-in from POLITO) and within NEWCOM IT++

6 Activities A campaign of simulations aimed at validating system oriented models from circuit level simulations A campaign of simulations aimed at comparing various system level simulation tools Methodology for extraction of system oriented models from measured data or standard HB (multitone) circuit simulations Comparison of MWO/VSS with MATLAB/Simulink, MHOMS and IT++ Quasi-memoryless models Models with memory (Wiener-like) Collaboration with Chalmers Univ. includes a PhD student exchange for the period April-June 2005 (maybe to be extended)

7 Quasi-memoryless models Model for narrowband modulated signal (Complex) envelope representation of input and output signals, envelope slowly varying vs. carrier: xt ( ) = Re xt ( )exp( jω t) = x ( t) cos ω t+ x ( t) { } ( ) c y() t = Re y ()exp( t jω t) = y ()cos t ω t+ y () t { } ( ) c Static envelope model yt () = G( xt () ) xt () G complex descriptive function identified with AM/AM AM/PM circuit simulations/measurements: c c

8 Example: a bipolar PA for 1900 MHz GSM systems MWO library example: integrated bipolar PA matching on MHz AM/AM and AM/PM extracted through single tone circuit simulation LS S 21 parameter CAP ID= C2 C= 1e4 pf RES ID= R1 R= 4000 Ohm RES ID= R2 R= 5300 O hm V_METER ID=VM1 CA P ID= C1 C= 1e4 pf DCVS ID= V1 V= BiasVolts V RES ID= R3 R= 225 Ohm TLIN ID= TL2 Z0= 50 Ohm PORT_PS1 EL= 49.5 Deg P= 1 F0= 1900 MHz Z= 50 Ohm PStart= -60 dbm PStop= 0 dbm PStep= 1 db CAP ID= C4 C= 0.91 pf V_METER ID=VM2 GBJT ID= GP1 1 B 2 3 C 4 S E RES ID= R4 R= 250 Ohm IND ID= L1 L= 9.34 nh RES ID= R5 R= 630 Ohm SUBCKT ID= S1 CAP NET= "Digital Phase S hifter" ID= C5 C= 1.27 pf 1 2 TL IN ID= TL1 Z0= 50 Ohm EL= 64 Deg F0= 1900 MHz PORT P= 2 Z= 50 Ohm IND ID= L2 L= 0.8 nh TLIN ID= TL3 Z0= 50 Ohm EL= 90 Deg F0= 2000 MHz CAP ID= C3 C= 1e4 pf

9 Example: quasi memory-less case AM/AM and AM/PM interpolated with fifth order polynomials as a function of absolute value of input power AM/AM AM/PM 10 AMtoAM[PORT_2,1] (dbm) AM_AM -40 AM_PM 0 DB(AMtoAM_LS[VSA.M2,1024,0]) (dbm) DB(AVGCH_INPDB[VSA.M2,1.9e9,2e6,0]) (dbm) -60 AMtoPM[PORT_2,1] (Deg) Ang(S21_LS[VSA.M2,1024,0]) (Deg) Power (dbm) Power (dbm)

10 Example: MWO/VSS The PA circuit schematic from MWO can be introduced into a system level simulation in VSS RND_D ID= A1 M= 2 RATE= 1/_TFRAME RNG= Auto QAM_TX ID= A2 M= 16 OUTLVL= Eb_N0 db OLVLTYP= Bit Energy (db) SYMRAT E= CTRFRQ= 5 G Hz PLSTYP= Rectangular ALPHA= 0 PLSLN= NL_S ID= S1 NET= "A" TP ID= 1 AWGN ID= A3 PWR= 0 db RCVR ID= A4 1 2 R D BER ID= 3 SWPVAR= Eb_N0 SWPTYP= Eb/N0 MNERR= 25 BER IQ 3 TP ID= RANDOM BIT GENERATOR QAM MODULATOR PA link to Circuit schematic AGWN QAM DEMODULATOR

11 Circuit Level System Level Link

12 Two-tone simulations Circuit level simulations (MWO) compared to system level simulations both in VSS and MATLAB Simulink VSS (dbm) Model 0 Matlab (dbm) Circuit HB (dbm) -50 Spectrum Input tones -20 dbm (1dB input) Space tones 2MHz Frequency (MHz)

13 System level simulators: VSS/MATLAB Simulink 16 QAM modulated input at 20 dbm power level VSS MATLAB

14 System level simulators: VSS/MATLAB Simulink BER 16 QAM System with 1900MHz Amplifier

15 System level simulators: VSS/ MHOMS Same kind of comparison with in-house system simulator shows the effects of nonlinearities in the case of strong compression P OUT BER P IN

16 System level simulators: VSS/IT++ A first consistency analysis is carried out without PA (only AWGN) with a QPSK modulated input QPSK_TX ID= A2 OUTLVL= Eb_N0 db OLVLTYP= Bit Energy (db) RND_D SYMRATE= ID= A1 CTRFRQ= 5 GHz M= 2 PLSTYP= Rectangular RATE= 1/_TFRAME ALPHA= 0 RNG= Auto PLSLN= TP ID= 1 Eb_N0 = sweep(stepped(0,7,1)) AWGN ID= A3 PWR= 0 db RCVR ID= A4 BER ID= 3 SWPVAR= Eb_N0 SWPTYP= Eb/N0 MNERR= 25 BER 1 2 R D IQ 3 TP ID= 2 5 4

17 System level simulators: VSS/IT++ Consistency analysis: BER BER confirms erfc behavior

18 System level simulators: VSS/IT++ PA nonlinear effects: QPSK modulated PA 1 db compression point reference is BER w/o PA PA nonlinear effects negligible in this example IT++ compares very well with VSS 1 million simulated bits: simulation time drops from 2 min. in VSS to 10 sec. in IT++

19 Wiener-like memory-model Linear with memory + nonlinear memory-less model x() t Polynomial NL The Wiener like model is obtained from the Volterra time domain representation, by using discrete time M LTI P M + y( t) ( ) ( ) ( ) ( ) ( ) n = 1 i n i + p i i n j 1 p p 1 p i= 0 p= 2 i1= 0 ip = 0 j= i1 y t h t x t h t t x t M neglecting memory in nonlinear part i LINEAR WITH MEMORY NONLINEAR WITH MEMORY

20 Model identification I In order to identify the two blocks the following procedure is followed With at least 10 db backoff from PA compression a two tones circuit simulation is performed M ( ) ( ) ( ) y t = h t x t n 1 i n i i= 0 ( ) ( ) h1 t1 h1 t M LINEAR WITH MEMORY IDENTIFIED THROUGH LEAST SQUARE OPTIMIZATION Memory M corresponds to h 1 (t M+1 ) negligible

21 Model identification II At 1 db from PA compression a two tones circuit simulation is performed The nonlinear part is obtained by subtracting the linear model already identified ( ) ( ) ( ) ( ) e t y t h t x t An L-order polynomial fit as a function of the input sample amplitude (input drive) is extracted M = n n 1 i n i i= 0 L ( ) ( ) e t = E x t n k n k=1 k

22 Simulation results Two tone input with 10dB backoff

23 Simulation results Two tones at 1dB point

24 Acknowledgements The presentation includes work from colleagues from the Microwave Measurements Group: Prof. Andrea Ferrero Dr. Valeria Teppati