Migrating an Academic DSP Lab from the C30 to the C67 EVM

1
Migrating an Academic DSP Lab from the C30 to the
C67 EVM

• by Keith Hoover,
• Dept. of Electrical
• and Computer Engineering,
• Rose-Hulman Institute of Technology,
• Terre Haute, Indiana
• Keith.E.Hoover_at_Rose-Hulman.Edu

2
Goals of this Presentation

• Discuss pros and cons of migrating academic DSP
  lab from C30 to C67.
• Briefly describe 3 RHIT courses that use C67.
• Describe the C67 DSP Projects course in detail.
• All C67 EVM DSP Projects course handouts are
  available for download at www.rose-hulman.edu/
  hoover/

3
The Thrill of RAW Power...

• Who could resist the chance to upgrade their DSP
  lab from an old 60 MIPS TMS320C30 DSP board to a
  new 1600 MIPS C67 EVM?
• After all. the C67 EVM is not much more expensive
  than the C30!
• TMS320C3X EVM 995 (List), 695 (Univ.)
• TMS320C6701 EVM Bundled with CCS EVM tools
  2495 (List), 1495 (Univ.)
• Source of EVM Prices www.ti.com/sc/docs/gen
  eral/dsp/programs/toolweb.htm

4
Bottom Line

• 27X increase in raw computing power, with only
  2.2X increase in cost!
• But thats not all...
• Consider the added bonus of having much more
  on-board RAM, and a really powerful integrated
  code development environment (Code Composer
  Studio!) bundled with the C67 EVM!

5
Code Composer Studio IDE

• CCSs incredibly rich features include
• Code Editor
• Code Gen C6x Compiler/Assembler/Linker
• Debugger Download/Single Step/Breakpoint
  Watch Window, host console I/O
  code profiling, simulator, graphing.
• DSP/BIOS Probe/Trace/Monitor/Data Streams
• RTDX (Real time data exchange) High-level host
  interface exchanges data with standard PC
  applications, such as LabView or Excel.
  

6
Are you sold on upgrading your DSP Lab to the
C67 EVM?

• A WORD OF CAUTION you need to plan CAREFULLY
  before you take the C67 plunge!
• Compared with our C30 boards, there is so much
  more for us to learn! And we are still learning
  after one year.

7
Our Original C30 DSP Project Course was able to
Teach it all in one 10-week Quarter,
including....

• C30 DSP Architecture
• Assembly Language Programming
• C30 C Language Programming
• Analog I/O Interrupts DMA
• PC/DSP Communication
• High-level DSP tools (DSPWorks, QED Filter
  design, code generator)
• Weekly Projects Final Term Project.
• C30 DSP Boards from 3rd Party Mfr. (These 6-year
  old C30 boards are no longer manufactured or
  serviced.)

8
But a price had to be paid for making the raw
power of the exciting, state-of-the-art C67 DSP
hardware and the sophisticated features of the
bundled Code Composer IDE available to our
students

• The price we paid was not economic.
• It was one of information overload!
• No longer could we successfully teach it all in
  a single 10-week course!

9
Teaching the C67 EVM is like flies on an
elephant...

• We are the flies.
• The C67 is the elephant!

10
None of us flies can possibly see the whole
elephant at once. Depending on where we land on
the elephant, each of us can get a very different
idea of what the elephant is all about!
11
Some challenges presented in migrating from C30
to C67....
12
Example 1. Manuals needed to Program C30 EVM
13
A few of the Manuals Needed to Program the C67
EVM Board! (This is NOT all of them!)
14
Then you need a book that tells you what the data
books say!

• C6x-Based Digital Signal Processing
• by Nasser Kehtarnavas
• Burt Simsek
• Prentice Hall 2000

15

• Example 2.
• C30 board communicated with PC via simple
  memory-mapped pass-through registers.
• C67 EVM can communicate with PC via a much more
  flexible (that means more complicated!) AMCC
  Matchmaker PCI controller chip, this chip
  transfers data pass-through registers, or
  mailbox registers, or FIFO registers.

16
Example 3. The C67 EVMs stereo A/D and D/A
converters on a flexible Crystal Semiconductor
CODEC chip, ---which requires a 76-page data
manual!
17
We solved the C67 information overload problem
by breaking our C67 instruction into the
following three, somewhat independent, 10-week
(1-quarter) courses

• EC332 Computer Architecture II (Yoder)
• C67 good example of modern parallel pipelined,
  VLIW CPU architecture.
• C67 assembly language programming
• Assembly optimization
• Software pipelining
• Writing optimized C code
• Use of Code Composer Studio (simulation)
• Based on TIs C62xx DSP Design Workshop course
  notes.

18

• EC597 Real-Time Systems Programming (Yoder)
• Patterned after TIs Real-Time Software Design
  Workshop course notes.
• Focuses on real-time programming concepts that
  promote more efficient real-time DSP programs,
  such as double-buffering, event scheduling, data
  profiling, DSP/BIOS I/O and data analysis, RTDX
  (real-time date exchange) capabilities.
• Does NOT dwell on the signal processing
  algorithms themselves.
• Significant term project required. (Ex broom
  balancer, MPEGII audio decoder, phased microphone
  array, image recognizer, internet telephone.)

19
EC581 DSP Projects (Hoover)

• DSP Theory Class is prerequisite.
• Focuses on real-time implementation of common DSP
  algorithms previously studied in DSP theory
  course.
• No Assembly Required! (Only C67 C-language and
  MSVC are used.)
• Independent of other two C67 classes, thus the
  emphasis is on writing simple, working C code
  rather than on writing optimal code.

20
EC581 (DSP Projects) consists of three
components...

• Nine single-week projects (2 person teams)
• Classroom demonstrations of selected topics of
  interest
• Term project.
• All reports are memo-style with all
  programs, data, and data analyses included as
  attachments.

21
EC581 Meetings

• Two lecture hours per week (to discuss the weeks
  lab and demonstrate advanced topics)
• One 3-hour scheduled DSP lab per week.
• Students expected to work in DSP lab on a walk-in
  basis for approximately six additional hours per
  week.

22
List of Weekly DSP Lab Projects
Each of these lab handouts (MS Word 97) may be
downloaded at
http//www.rose-hulman.edu/hoover/

• 1. Code Composer familiarization. Audio
  sampling, study of effects of quantization on
  speech recognition, reverberation, comb filter,
  flanger
• 2. FIR filter design (MATLAB). Real-time
  floating and fixed-point FIR filter
  implementation.
• 3. IIR filter design (QEDesign). Real-time
  implementation. Implementation of Wah-Wah audio
  sound effect.

23

• 4. Digital automatic gain control with silence
  threshold.
• 5. Audio VU meter with separate target and host C
  programs communicating through PCI interface via
  FIFO register.
• 6. Radix-4 FFT real-time audio spectrum analyzer.
• 7. MATLAB narrow-band noise reduction via
  adaptive filtering. (Once MATLAB LMS adaptive
  filter works, implement in real-time on C67 EVM.)

24

• 8. Use of Hyperception RIDE 4.2 to perform
  real-time C67 EVM-based digital filtering and
  audio scrambling and descrambling via spectral
  inversion.
• 9. DSP Scavenger Hunt - Use RIDE 4.2 digital
  filtering and LMS noise reduction to reduce noise
  level in various noise-polluted audio clips that
  indicate the location of 5.00 bills hidden
  around campus! First come, first served!

25

• 10. Term Project (to be chosen by student)
• Example Projects Done in Past
• Real-time audio spectrum analyzer
• Ultrasonic chirp radar
• Touch-tone DTMF decoder
• Guitar Tuner
• Voice-operated security lock
• Underwater voice communication system using
  ultrasonic-carrier
• Guitar effects box

26
Special Class Demos

• M.S. Students describe DSP their DSP-related
  thesis work.
• Present theory of Linear Predictive Coding and
  show EVMC30 board 2400 bps LPC demo. With pitch
  shifting, whisper effects!
• We need this wonderful C30 LPC demo ported to
  C67!!

27
Pros of migrating from C30 to C67 EVM-based DSP
Projects

• Students more motivated to work with
  state-of-the-art DSP.
• Code Composer Studio offers superior source-level
  debugging environment.
• Giving the students the sampling template
  program to work from allows them to concentrate
  on DSP algorithm development and not get lost in
  the forest (details of CODEC and McBSP serial
  port configuration) for the trees!

28

• Real-time DSP algorithms not as time critical
  due to the faster speed of C67.
• C67 EVM will most likely remain supported for
  longer time than if we had bought new C30 boards
  at this time.
• New DSP laboratory textbooks will probably focus
  on the C67, not the C30.
• Dealing with C67 EVM complexity is probably good
  preparation for the technological challenges that
  lie in the future. (These NOT getting any
  simpler!)

29
Cons of migrating from C30 to C67

• INFORMATION OVERLOAD! Need for several C67
  classes. Learning time (for students AND for
  instructors) is long compared to C30.
• Increased reliance on soft manuals in .PDF --
  hard to work with (though free to students!)
• Code Composer does not appear to be stable under
  Win 98 -- we could only Win 95 or NT!
• As an EE, I dont like getting so far away from
  the hardware level.

30

• C67 EVM communicates with PC running Microsoft
  Visual C.
• Just as going from the C30 to the C67 has become
  an order of magnitude more complicated, so has
  Microsoft C become an order of magnitude more
  complex.
• With the latest version of MSVC, it is a majore
  challenge to write anything more than a simple
  MSVC console application.

31

• Conclusion
• The C67 DSP board has successfully provided our
• students with three different and valuable
• kinds of lab experiences in three different
  courses
• Low-level Architecture and Assembly Programming
• Mid-level DSP Project Applications
• High-level Real-time Programming Issues
• Though not without its frustrations, we realize
• that no road that leads into the future is
  smooth.

32
We feel that the C67 EVM will be an excellent
vehicle with which we can ride into the future
with our students over the next five years!
33
(No Transcript)
34
(No Transcript)
35
(No Transcript)