DAC Presentation kit

1
(No Transcript)
2
Matlab as a Design Environment for Wireless ASIC
Design
Erik Lindskog
Beceem Communications, Inc.
elindskog_at_beceem.com
June 16, 2005
3
Introduction to Beceem

• Wireless communication system developer
• Using Matlab for ASIC design
• Product development phase
• Algorithm development - Matlab
• Algorithm and system modeling - Matlab
• RTL design Verilog
• RTL verification - SystemC, Verilog and Matlab
• SW design C
• SW verification C and Matlab

4
Advantages With Matlab

• Rich support for mathematical operations and
  visualization
• Most common tool for signal processing research
• Graduates in communication and signal processing
  engineering are typically highly familiar with
  Matlab
• Fewer people have experience in programming in
  C
• Full fledged programming as well as scripting
  language
• Interpretating language Integrates well with
  other programs
• Powerful debugging support
• Not a typed language - Highly moldable

5
Matlab Simulator Design

• Common simulator for
• PHY algorithm development and studies
• Small network full PHY studies of
• Effect of co- and adjacent channel interference
  on PHY algorithms
• Synchronization and hand-over
• Link-adaptation
• Fixed point modeling
• RTL signal processing reference
• Standalone code for
• Independent lower complexity studies
• Generation of independent references

6
Matlab Simulator Design, cont.

• Organization
• Common core source code
• Simulation examples driving source code
• Single file with default parameter settings
  enables developers to work reasonably
  independently while using the common code

7
Open Source/Freeware Software

• CVS - Concurrent Version Control system
• Cygwin Unix emulator for Windows
• CVS Tortoise File browser based CVS client
• Integrates external free software tools like
  ExamDiff for CVS diffing
• WinCVS GUI based CVS client

8
Signal Processing Design Flow

• Module design
• In standalone code, or
• Direct in main simulator
• Floating point system modeling
• Integration of modules into main simulator
• Floating point performance simulations

9
Signal Processing Design Flow, cont.

• Fixed point module design
• In module design environment to excite corner
  cases
• Integrated in main simulator to get realistic
  signal levels
• Fixed point system modeling
• Full system fixed performance simulations
• Ability to turn fixed point on/off per block
• Analysis of quantization statistics to pin-point
  problems

10
Matlab RTL Design Flow

• Matlab fixed point design handed over to RTL
• Module verification
• File-in-file-out test vector comparison between
  MATLAB and Verilog
• Extensive testing to cover corner cases in fixed
  point arithmetic
• Random signals to guarantee fixed point
  equivalence

11
Matlab RTL Design Flow

• System testing
• Side-by-side testing of Verilog and MATLAB in
  SystemC testbench
• MATLAB simulator compiled and linked to C-program
  called from SystemC testbench
• Verilog code excited from SystemC testbench
• Verilog simulation speed is the limiting factor
• Common test for Matlab and Verilog code
• Directed testing for initial coverage
• Random testing for extensive coverage
• All system tests part of Matlab automated
  regression testing

12
Matlab SW Design Flow

• Matlab fixed point design handed over to SW
• Module verification
• File-in-file-out test vector comparison between
  MATLAB and C SW code
• Simulation example test cases in Matlab for C
  SW code
• System testing
• Side-by-side testing of SW on FPGA and Matlab
  system simulation

13
Fixed Point Modeling

• Currently performed in regular Matlab
• Regular arithmetic operations where precision is
  allowed to grow
• Quantization with quantization functions where
  precision is to be limited
• Automatic fixed point statistics collection

14
Object Oriented Fixed Point Modeling

• Simultaneous floating and fixed point
• Fixed point format embedded in variable
• Matlab object oriented support limited
• Assignment operator cannot be overloaded
  Traditional quantization with assignment
  operation in C cannot be done
• Function overloading only sensitive to first
  argument
• Catalytic have an object oriented fixed point
  modeling tool
• Matlab also now have an object oriented fixed
  point toolbox

15
Fixed Point Analysis

• Analysis of fixed point quantization statistics
• Headroom E.g. headroom above 67th percentile
• Signal to quantization noise ratio SQNR
• Over- and underflow ratios
• Signal level histogram
• Full system simulation fixed point analysis
• Example Matlab implementation
• Fixed point analysis function
• Example call fxp_stat(x, 2, 5)
• Store statistics in function call stack tree
  structure
• Analyze statistics after completed simulation
• Block-swapping analysis
• Switching of fixed and floating point blocks

16
Lab use of Matlab Simulator

• Modeling of Tx or Rx part of modem against HW Rx
  or Tx modem
• Interoperability testing with signal generators
  and signal analyzers
• MAC-PHY interoperability testing
• Simple RF testing
• Over the air transmission demonstration when
  combined with RF

17
Matlab Regression Testing

• Developer regression testing
• To many tests to run all
• Directed and/or randomly selected set of tests.
• Automated Regression Testing
• All regression tests checked for each block of
  commits
• Breaking commit identified for each failing
  regression test
• Result of testing automatically emailed to
  developer
• High SNR Tx-Rx tests
• Single SNR BER tests
• Simulation example execution tests
• RTL verification tests mimicing exact use in RTL
  test bench

18
Matlab Regression Testing, cont.

• Automated Random Testing
• Script generating random traffic for testing of
  Matlab simulator and RTL
• Periodic random testing of code in repository
  with random Tx-Rx tests
• Result automatically emailed to developers
• Provides extensive coverage
• Scripting language - Matlab

19
Speeding up Matlab Code

• Scalar loops
• Loop over scalar operation are slow
• Vectorization
• Convert scalar loops to matrix operations
• Conversion to C-code
• Compilation of Matlab
• Removes the need for other language
• Limits the need for vectorization of code
• Limits duplication of code in Matlab and C
• Example compiler Catalytic Matlab Compiler

20
Summary

• Matlab
• Powerful tool for signal processing design
• Lessens distance between algorithms and HW
• Facilitates for algorithm designers to be closely
  involved in fixed point testing and RTL
  verification
• Interacts easily with other programs and
  equipment
• Attractive design environment for ASIC design