Design and Verification of High Efficiency Power Amplifier Systems Sean Lynch Platform Engineering Manager MATLAB EXPO 2013 1
What is Nujira? Nujira makes Envelope Tracking Modulators that make power amplifiers more efficient. We are based in Cambridge (UK) Nujira was 10 years old last year We have over 70 employees, 34 of which have MATLAB licenses. We have over 200 patents covering Envelope Tracking Technology We have developed three major product lines: Low Power 1 Watt High Power 20 Watt Ultra Power KWatt 2
Contents This presentation will cover the following: Envelope Tracking Principles Closing The Design Loop System Level Modelling RF Test System PA Characterisation Envelope Tracking IC validation Flexible Development Platform Development Summary 3
ENVELOPE TRACKING PRINCIPLES 4
Envelope Tracking Principles Envelope tracking was first described over 60 years ago. Envelope tracking describes an approach to RF amplifier design in which the power supply voltage applied to the power amplifier is constantly adjusted to ensure that the amplifier is operating at peak efficiency for the given instantaneous output power requirements. Nujira was the first company to produce a manufacturable implementation with our.h product line. 5
Envelope Tracking Principles RF V supply RF PA Antenna Fixed Supply Envelope Tracking (ET) 6
Envelope Tracking Principles To achieve the best system efficiency, we need to determine the optimum V supply for each output power level. We call the mapping function between Pout and V supply a Shaping Table Device Peak Power V supply (Volts) Power Amplifier Operating Range Instantaneous Input Power (dbm) 7
CLOSING THE DESIGN LOOP 8
Closing The Design Loop We have a system level based development flow: Model Design Build RF Transceiver Verify The same approach is used in the development of Envelope Tracking Integrated Circuits (ETIC) and internal test equipment: Envelope Tracking IC Model Design Build Flexible Development Platform Verify 9
SYSTEM LEVEL MODELLING 10
System Level Modelling We model the behaviour of key components to predict full system performance. The PA model is based on data obtained using tools we have developed internally (PA Characterisation) The original PA models were developed using ADS tools, but porting them to MATLAB resulted in quicker tools with greater dynamic range. We use different modelling techniques for each of our Modulator product ranges. 11
System Level Modelling For the high and ultra power Modulators we have used Simulink to model the modulator behaviour. These high level models have enabled us to optimise component values without excessive bench time. For the ET Integrated Circuit Modulators we use the Cadence design tools backed up with MATLAB functional models. The MATLAB models provide much quicker results than the pure IC based tools. The following MATLAB toolboxes are used in the system level simulations: Signal Processing Communication Systems DSP System RF 12
RF TEST SYSTEM 13
RF Test System - Introduction This RF System Test Bench is used to: Characterise PA behaviour For use in system models and creating Shaping Tables Measure system level parameters: Efficiency, ACP, EVM, noise, tracking accuracy For use in system and ETIC validation RF Test System PA Characterisation 14
RF Test System Block Diagram The following shows the key software and hardware components: PC running MATLAB USB ATE Control Manual GUI Power Meter Power Meter Waveform Generation Performance Analysis Adaptation Algorithms (DPD, Timing, AGC, QMC) Modulator Drivers Instrument Drivers USB Nujira Flexible Development Platform RF Envelope RF Nujira Modulator PA Supply Voltage/ Current RF PA RF Load Baseband Driver (Mexw32/64) Instrument Signal Processing Toolbox GPIB DMM DMM Power Power Supply Supply Oscilloscope Spectrum Analyser 15
RF Test System - Measurements The key to characterising PAs for ET is instantaneous measurements of power, current and voltage. We take instantaneous measurements at four points in the system. (1)+(2) PA Gain/Phase (3)+(4) Split Power Baseband ADC#1 Baseband ADC#2 (1) RF Input (2) RF Output RF Envelope Modulator V supply PA RF (3) Voltage (4) Current Oscilloscope Channel#1 Oscilloscope Channel#2 16
RF Test System Evolving Requirements In the last year system complexity has increased considerably So we have produced one of the first multi band MIPI RF front end reference designs PA PA MIPI Bus Switch RFIC PA Sensors ETIC PA 17
RF Test System- Automation Which has resulted in the following changes to the RF Test System: Added MIPI control for ETICs, PAs and switches. Added Multiple RF band measurement Added RF Noise measurement capabilities Band Select Switches 18
RF Test System - Applications The software can be run in the following modes: A compiled stand alone application for customers Using the MATLAB Compiler Manual control GUI for developers ATE mode for design verification Which has saved over $2.0M in manual test costs to date Manual GUI ATE GUI 19
RF Test System - Reports The automated test software records test results in a repeatable manner, storing data on the network and MySQL database. A new automated report generator enables faster data analysis. Equipment Measurements Summary Sheets Report Generator Detailed Results Files Results Plots 20
RF Test System - Calibration To ensure that repeatable and consistent results are obtained we have developed a series of RF bench calibration Wizards These are used to calibrate: Oscilloscope measurement points for: Modulator voltage and Modulator current sense RF power measurement points across frequency Analogue envelope voltage levels 21
PA CHARACTERISATION 22
ET Surface Explorer ET Surface Explorer is a extension to the RF Test System that takes measurements across a range of PA operating points using a patented measurement technique. This enables us to create a surface describing power amplifier gain and phase as a function of V supply and input power Large numbers of operating points V supply (Volts) ET Surface Explorer GUI Instantaneous Input Power (dbm) 23
PA Vsupply [Volts] PA Gain [db] ET Surface Explorer - PA Surface The following shows a captured PA surface, with an ISO gain Shaping Table overlaid. PA Vin [Volts] 24
ET Surface Analyser Once we have the surface of a PA we can perform the following steps: Extract and manipulate Shaping Tables Predict system performance Compare predicted and actual performance Using the combination of ET Surface Explorer and ET Surface analyser we can characterise new PA in hours, not days. ET Surface Analyser GUI 25
ET Surface Analyser - Plots This tool enables you to visualise the behaviour of the PA under test: Supply Impedance Instantaneous Efficiency Phase Surface Gain Surface 26
Benefits of Envelope Tracking Fixed drain mode ET Mode Improved battery life ~30% Simplified cooling Requirements 27
Benefits of Envelope Tracking The -30% saving in battery power is dependant on signal statistics 28
Benefits of Envelope Tracking ET Surface Analyser results analysis: Red predicted Blue measured Higher Output Power Higher device peak power Larger RF cells More linear output without digital predistortion for handset applications Improved ACP Fixed Drain Envelope Tracking 29
ETIC VALIDATION TOOLS 30
ETIC Block Level Validation The block level test framework gives ETIC designers easy control of the following: ETIC MIPI Interface Envelope signal (FDP) Spectrum analyser Scanning DMM x40 channels Oscilloscope x4 channels Power Supplies x6 Temperature chamber Temperature Sensors I 2 C EEPROM calibration data Block tests can be combined to allow volumetric testing over temperature and other operating parameters 32
ETIC Block Level Validation Test limits are derived from the device simulations are saved with the results to enable easier analysis: Result.iTestA473 errampa_stg2, 1/40 mirror of pmos output device current 91.70 119.00 146.30 µa Result.iTestB473 errampa_stg2, 1/40 mirror of nmos output device current 86.50 116.80 147.10 µa Result.iTestA474-1 errampa_opstage, mirror of ab_bias setting current 11.34 12.42 13.50 µa Result.iTestA474-2 errampa_opstage, mirror of ab_bias setting current 22.97 24.65 26.33 µa Result.iTestA474-3 errampa_opstage, mirror of ab_bias setting current 34.45 36.79 39.13 µa A generic GUI enables access to the ~150 different block test results. 33
FLEXIBLE DEVELOPMENT PLATFORM DEVELOPMENT 34
TX Digital System Gain Flexible Development Platform - Requirements We have developed custom ET test equipment that has the following key features: Outputs: RF Frequency from 0.6 to 2.7 GHz RF Sample rate 245 MSPS 10 millisecond playback memory Modulator control Inputs Dual wideband receivers 0.6 to 2.7 GHz RX Sample rate 491 MSPS RF Configuration Algorithms ET Generation Interface (EGI) Digital Pre-Distortion (DPD) Dynamic QMC Correction Automatic power level control Normalisation Matlab TX Digital Back-Off Volts Frac Shaping Dither 2 Delay Table Functions U16 Software EGI IQ RF Dither 2 IQ Functions U16x2 TX Digital DPD DPD TX Equ Back-Off Correction Headroom Headroom Digital S D R A M 14 db Gain S D R A A M Max: -1.6dB Default: -3.6dB Min: -5.6dB TX Digital RF Gain Adjust TX Path Block Diagram Analogue Analogue Envelope RF A B TX RF 14 db TX RF Gain Low Gain Gain High C B C Def: 0.0 db Max: 0.0 db Max: 0.0 db Min: -31.0 db Min: -31.0 db 35
Flexible Development Platform Design Flow We have followed the development flow show below for the Baseband firmware. Currently we use manual methods to translate MATLAB into FPGA and embedded C code. But have evaluated automating the process using the C generation tools. The FPGA code and fixed point models are provided to customers, so they can integrate our IP into their solutions. Phase#1 Modelling Application [MATLAB] Phase#2 Development Application [MATLAB] Phase#3 Product Application [MATLAB] Algorithms [MATLAB] Baseband Configuration [MATLAB] USB Driver [Mex32/64] Algorithms [MATLAB Fixed Point] Baseband Configuration [DSP C ] USB Driver [Mex32/64] Algorithms [DSP C /FPGA] Baseband Configuration [DSP C ] USB Driver [Mex32/64] FDP Design Flow 36
Flexible Development Platform - Validation We were worried that small data sets may not fully verify some of the algorithms. So we structured the code so could run all three versions in parallel. This allowed verification over long time periods and operating conditions Live Capture Data Algorithms [MATLAB] Algorithms [MATLAB Fixed Point] Algorithms [DSP C /FPGA] Output Comparison Concurrent Validation 37
L 6 N 5 3 2 6 5 3 2 Flexible Development Platform - Manufacturing The Flexible Development Platform is a complex piece of test equipment with 28 ports that need production test and calibration We have used a programmable RF switch array to enable the automated testing of complete unit from MATLAB This includes calibrating TX and RX RF attenuators at 60 different RF frequencies 87104B 2 3 5 PA block A 6 2 87104B 87104B 6 Filters block 87104B 2 3 5 3 5 2 D5 5 6 C M 3 B 6 5 87104B F6 6 2 87104B 3 H Z11 Power Meter (USB) Scope Chan#1 (GPIB) 87104B Trigger to Scope 2 3 5 6 87104B G Signal Generator (GPIB) 2 87104B 87104B 6 3 5 M2 5 2 6 D K 3 87104B 5 Spec (GPIB) F 87104B 6 2 3 2 H6 A6 Agilent L4490A/91A RF Switch Platform 3 5 E 6 RF Switch Test Matrix 38
Summary Software Integration We have used the following 3 rd party integrations in our developments 32/64-bit MEX Functions Optimised C functions 3 rd Party DLL loading Aardvark I 2 C/SPI controller Pico Temperature Sensors Flexible Development System Driver File Access XML Files Excel Files Java/C# Libraries Encryption Tools Database Access 3 rd Application Execution Microsoft Visual Studio DLL compiler InnoSetup - Windows installer builder MATLAB Com Interface FDS control from C# & Labview 39
Summary MathWorks Tools We have used MATLAB tools throughout our development process from system modelling to device validation We have used the following MATLAB products at each stage: System Modelling Code Generation Simulink Fixed Point Designer Signal Processing Toolbox MATLAB Coder Communication Systems Toolbox DSP System Toolbox ETIC Testing RF Toolbox Instrument Control Toolbox RF Testing Signal Processing Toolbox Instrument Control Toolbox Customer Releases MATLAB Compiler MATLAB Builder NE It could be said that MATLAB is the glue that holds it all together 40
THANK YOU 41