Realtime Signal Processing

1
Real-time Signal Processing
1) System must keep up with incoming and/or
outgoing data 2) Delays between arrival of data
and its disposition must usually be limited
USB to PC
to 5V
Headphones
Microphone
2
Continuous Real-Time Processing
Data being acquired continuously. Data being
output continuously. Use DMA (direct memory
access) to allow data I/O without CPU
involvement. Output is continuous but
delayed. Can use buffers - one is filled
while one is processed while one is output.
3
Data Transfers by Direct Memory Access

• All data transfers have to be handled by the CPU.

Data are transferred directly to memory, enabling
the CPU to attend to other tasks.
4
Programming for Real-Time Systems
Keep algorithms fast and simple IIR filters have
lower order than FIR Use integer arithmetric
(e.g., 3-point smoother) And/or use hardware
enhancements DMA for data acquisition Hardware
multipliers to allow floating-point
arithmetic Pipelines All this with low power
consumption!
5
Processors in C6000 Family

• C67x floating-point processor.
• C64x fixed-point processor.

6
Texas Instruments TMS320CxDSP chip block diagram
Features of DSP chip
Multiple ALUs Floating point hardware Internal
memory for program and data DMA control External
interrupt controller External ports Timers
7
The C6000 Block Diagram
8
The C67x Data Paths

• The C67x CPU is divided into two parallel parts,
  Data path A and Data path B.
• It is stereo.
• With this architecture, it is possible to process
  both left and right channels of an audio signal
  at the same time.

9
TMS320C6713 DSK Hardware
C6713 CPU
Mic In
Line In
Headphones
Expansion
Codec
USB
5V Supply
LEDs
JTAG
RAM
Switches
10
Comparing Target Hardware
11
Current Methods to DeployEmbedded Real-Time
Systems
Design algorithms at high level (e.g.,
MATLAB) Select target hardware Cost Power
consumption Silicon area Development
time Translate algorithm to target
code Over-the-wall Lack of integrated
development (cross-compilers) Hand
optimization Implement
12
A Call for Change

• Cultural wall between Electrical Engineering
  and Computer Science
• CS deals with discrete mathematics
• Use computational models based on sequential
  machines
• Support non-deterministic changes in program
  control but abstract out physical realities such
  as notions of concurrency and physical
  constraints on computation
• EE deals with continuous mathematics
• Use analytical models based on transfer functions
• Explicitly recognize time but have difficulty
  with non-deterministic changes in program control
  and partial or incremental specifications
• Needs
• Bi-cultural engineers comfortable in both
  worlds
• Design models that integrate analytical and
  computational approaches
• Methods to define and design robust systems
• See The Discipline of Embedded System Design,
  Thomas A. Henzinger and Joseph
• Sifakis, Computer Magazine, October 2007, pp.
  32-40 (IEEE Computer Society).

13
Generating target code from MATLAB

• MATLAB compiler
• mcc m yourmfile recorder player2
• generates C-source files, yourmfile.exe
• and yourmfile.ctf (environment) file
• Simulink
• Uses block diagram representation of a system
• generates C-source files and yourmfile.exe
• Vehicle for MATLAB support of
• external hardware
• Real Time Workshop
• Real Time Windows Target
• Licenses required
• Based on S-Functions
• Generates downloadable code

14
Simulink model for thecar cruise control
15
Simulink model for theautomatic gain control
Read input from and write output to workspace
agc.mdl
Uses DSP Blockset and Real Time Windows Target
modules
16
Simulink model for theautomatic gain control
Read input from and write output to disk files
agc_c.mdl
Uses DSP Blockset and Real Time Windows Target
modules
17
Simulink model for theautomatic gain control
Use real-time data input and output
agc_rt.mdl
Uses DSP Blockset and Real Time Windows Target
modules
18
Downloading files for Lab 6

• Data (MATLAB files) refspeed.txt (mat format)
  agc.txt (mdl format)
• Note instructions on website for using files in
  MATLAB
• As indicated, refspeed is in .mat format. You can
  force MATLAB to ignore the extension and load it
  as a .mat file using the command load mat
  pendin.txt
• To use the agc.txt and other model files for
  Simulink, you can change the extension to mdl.
• One way to do this is to use the dos rename
  command at the MATLAB command prompt
• dos('rename agc.dat agc.mdl')
• Note that Windows does not let you change the
  extension.

19
Notes on using mcc and RTW

• To deploy a compiler-generated stand-alone
  application to another machine, MATLAB Component
  Runtime (MCR) must be installed and configured on
  the target. Both the .exe and .ctf (Component
  Technology File) files must be on the target.
  Windows extracts a real-time operating
  environment from the .ctf file, which is platform
  dependent and apparently needs to be compatible
  with the source machine MATLAB version.
• Problems can be introduced if you change the name
  of the .exe file from mcc. The safest approach
  seems to be to let it keep the same name as the
  original m-file.
• RTW generates a single file that can be moved
  between MATLAB versions at least we havent
  encountered problems yet.
• An executable file can be built from agc_rt.mdl
  but not on the lab machines because supportable
  A/D and D/A modules are not available in the lab.

20
Automatic gain control
agcin.mat
Input and output files for agc_c.exe obtained on
DSP lab computer November 3, 2008
agcout.mat (Note that agc_c doesn't do much. It
just creates a new file in your default
directory, but this file contains your output for
the input agcin.)
21
Running automatic gain control from RTW
Using model agc_c.mdl (compiled April 1, 2008)
Using model agc_c.mdl (compiled February 16, 2006)
22
mcc compiler issue

• My compilation test produced this error message
• mccExcludedFiles.txt
• This file contains the list of various toolbox
  functions that are not
• included in the CTF file. An error will be shown
  if any of these functions are called at run-time.
  Some of these functions may be from toolboxes
  that you are not using in your application. The
  reason for this is that these toolboxes have
  overloaded some methods that are called by your
  code.
• If you know which toolboxes are being used by
  your code, you can use the -p flag with the -N
  flag to list these toolboxes explicitly. This
  will cause MATLAB Compiler to only look for
  functions in the specified toolbox directories in
  addition to the MATLAB directories. Refer to the
  MCC documentation for more information on this.
• 0 item(s) removed. You can review the
  mccExcludedFiles.log file for the list of
  functions not included in the CTF archive.

23
Final Project

• Objective
• Design a laboratory exercise that would be
  appropriate for a one or two week assignment for
  1563. It can be either a set of exercises or a
  design problem. Your report should include a lab
  write-up to describe the lab to students,
  including a background section on the area being
  studied
• Project Ideas (You are free to design your own.)
• Long-term spectrograms.
• Spread Spectrum Communication.
• (See Notes on Modulation on the website.)
• Effects of filtering on music and/or speech.
• 4. A simple speech recognition system.
• 5. Echo Cancellation.
• 6. Frequency transposition of speech.
• 7. The "cocktail party effect".
• 8. Brightness and contrast in color images.
• 9. Segmenting an image.
• Feature extraction in MRI images.
• Morphing instruments

24
Final Project Presentations

• Experiment 6
• Deploying MATLAB Algorithms
• Due Monday, April 6
• Exam 2 Wednesday, April 1
• Final Project
• Project titles due April 1
•     Class presentations April 13, 15
•     Final report noon, April 20