Distributed Simulation of Modern Communication Systems Using the Global Grid Exchange

1
Distributed Simulation of Modern Communication
Systems Using the Global Grid Exchange
June 1, 2006 June 1, 2006 June 1, 2006
Matthew Valenti Lane Dept. of CSEE West Virginia University mvalenti_at_wvu.edu Steve Armentrout Parabon Computation, Inc. Reston, VA steve_at_parabon.com Amy Beaudry Global Grid Exchange Fairmont, WV amybeaudry_at_wvhtf.org
2
Wireless is Big Business

• Over 1 Billion wireless phone subscribers
  worldwide.
• 205,829,280 in the US.
• About 200,000 cellular base stations (towers) in
  US.
• US Revenues of over 100 Billion per year.

3
Simulation of Communication Systems

• Before any new system is fielded, it must be
  extensively tested.
• Testing of proof-of-concept systems starts with
  computer simulation.
• Example Bit-level simulation of the 3-G wireless
  system UMTS/WCDMA.

4
A Simple Simulation

• Channel includes random effects of
• Noise (primarily).
• Fading.
• Interference.
• During each trial
• A packet or random data is generated and passed
  through system.
• Random noise is generated, usually according to
  Gaussian distribution

Error Counter
Random Data Input
Estimated Data
Modulator
Demodulator
Communication Channel
5
Raw Bit Error Rate of Several Modulations
6
A More Sophisticated Simulation
Random Data Input

• Channel code adds redundancy in the form of
  parity bits.
• Redundancy is used to correct errors.
• Decoder significantly increases computational
  complexity of the system.

Error Counter
Estimated Data
Encoder
Decoder
Modulator
Demodulator
Communication Channel
7
Simulation of IEEE 802.11g
8
Simulation Run Times
9
Simulation Run Times
10
Coded Modulation Library

• Coded Modulation Library (CML)
• Developed at WVU.
• Runs in matlab.
• Free software (licensed under lesser GPL).
• Features
• Modulation PSK, QAM, APSK, FSK.
• Coding convolutional, turbo, LDPC.
• Information theoretic bounds (channel capacity
  outage probability)
• Standards
• Cellular WCDMA, HSDPA, cdma2000
• Wireless LAN/MAN 802.11a/g, 802.16 (Wimax)
• Satellite DVB-RCS, DVB-S2
• Download
• Google keyword turbo codes, 2nd hit.
• http//www.csee.wvu.edu/mvalenti/turbo

11

• Initiative of WVHTC Foundation
• Only statewide public computing grid
• Desktop access to 1000s of computers
• WVU is one of the largest providers

12
Operation
Launch and Listen Local vs. Remote Exception
handling
13
Operation
Java native code Checkpoint/restart Prioritizati
on
14
Operation
Multi-platform MatLab extensible Highly
configurable
15
Configurations
16
Configurations
Internet
17
Configurations
18
Configurations
19
Architecture
CML
20
Developing Grid Jobs in Matlab

• Goals
• Work in matlab environment.
• Use same calling syntax as stand-alone CML
  library.
• Develop code in matlab, rather than java.
• Dont want to require a matlab license on each
  grid node.
• Implementation
• Use matlab compiler to create stand-alone
  executables.
• Considered to be native code.
• Enable cluster to run native code.
• One grid node used per simulation scenario.
• Send the compiled code plus a data file to the
  grid node.

21
System Requirements and Setup

• End Computer
• Java RE 5 and SDK
• Matlab 7
• Compiler needed if you want to make changes to
  the code.
• CML 1.5 or above
• http//www.iterativesolutions.com
• Global Grid Exchange software and account
• http//www.globalgridexchange.com
• Grid
• Windows or linux.
• Nodes must be on the Global Grid.
• Must be set to allow native code.
• Either matlab 7 or matlab runtime environment

22
Job Controller
23
4
7 tasks ran in parallel on the grid 2 were slower
than the local machine 5 were faster After 150
minutes, the local computer executed 159,013
trials, while the grid executed 1,408,483, nearly
an order of magnitude improvement.
3.5
3
2.5
Computations relative to 1.2 GHz P3
2
1.5
1
Dotted black line shows performance of local
laptop, a 1.2 GHz PIII w/ 512 Mbytes RAM, which
processes 64,140 simulation trials per hour.
0.5
0
0
0.5
1
1.5
2
2.5
Time in hours
24
14
11 tasks running in parallel 1 was faster than
the local machine (gold line) 9 were slightly
slower 1 was significantly slower (red
line) After 9.5 hours, the grid executed
6,019,410, nearly an order of magnitude
improvement over running locally.
12
10
8
Computations relative to 1.2 GHz P3
6
4
Dotted black line shows performance of local
laptop, a 1.2 GHz PIII w/ 512 Mbytes RAM, which
processes 64,140 simulation trials per hour.
2
0
0
1
2
3
4
5
6
7
8
9
10
Time in hours
25
40
11 tasks running in parallel After 24 hours, the
grid executed 16,630,510 trials an order of
magnitude improvement over running locally.
35
30
25
Computations relative to 1.2 GHz P3
20
15
10
5
0
0
5
10
15
20
25
Time in hours
26
Monte Carlo Optimization

• With the ability to run large numbers of
  simulations in parallel, it is possible to
  determine optimum system parameters via Monte
  Carlo simulation.
• Example FSK modulation.
• Used in most military communication systems.
• Also used in GSM cell phones and Bluetooth
  devices.
• Several parameters to optimize
• Modulation order (number of frequencies)
• h0 to 1 in 0.01 increments
• Modulation index (frequency spacing)
• M2, 4, 8,or 16
• Code rate (amount of redundancy in error control
  code).
• Rate is determined by h, M and amount of
  available bandwidth

27
2
1.8
1.6
1.4
1.2
1
Rate
0.8
0.6
CM Capacity of 4-FSK h0.5 in Fading
0.4
0.2
0
5
10
15
20
25
30
35
40
Eb/No in dB
28
(No Transcript)
29
Credits

• WVU Graduate Student
• Raja Katuri
• Parabon Programmer/Engineer
• Jim OConnor
• Parabon Systems Staff
• Mario Bulhoes
• Dabe Murphy
• WVU/LCSEE Systems Staff
• David Krovich
• Marc Seery
• The WVU portion of this project was accomplished
  without federal funding.