Performance Comparison

1
Adaptive Data-rate Controlled OFDM Demodulator
System (Real-rate Implementation)
CS6235 Final Project Jaehwan Lee Kyeong Keol
Ryu 23 April, 2001
2
Outline

• Whats the OFDM?
• Real-rate implementation
• System configuration diagram
• OFDM modulator and demodulator block
• MPI processing
• Demo. environment
• Adaptive feedback control for data transmission
• Experimental Environment
• Performance comparisons in MPI processing,
  feedback control
• Summary

3
Whats the OFDM? (1)

• Orthogonal Frequency Division Multiplexing (OFDM)
• A kind of wireless communication technique.
• Algorithm
• Algorithm development in 1970
• Hard to implement for real-rate because of short
  computing power.
• Characteristics
• Multi-carrier modulation and demodulation
• Avoiding inter-symbol interference using cyclic
  packet structure

4
Whats the OFDM? (2)

• Basic Structure
• Specification
• 128 block size, 32 guard signal, 64QAM symbol
  mapping

IFFT
Add cyclic prefix
Remove cyclic prefix
FFT
channel
Cn,0
Cn,0
Cn,1
Cn,1
Cn,N
Cn,N
Transmitter
Receiver
128 samples
32 samples
160 samples
5
Real-rate Implementation (1)

• System Configuration Diagram

Linux Machine Cluster Jedi1Jedi9
Linux Machine Marlet
Windows Machine
1
Network
Network
2
Application Demo. Environment
Data Generation
0
0
3

OFDM Modulator
8
OFDM Demodulator Using MPI
OFDM Receiver
OFDM Transmitter
6
Real-rate Implementation (2)

• OFDM Modulator and Demodulator Diagram

Modulator
IFFT
Mapping Symbols
P/S Conv.
Cn,0
Adding Cyclic Prefix
Cn,1
Cn,N
Network
Demodulator 1
FFT
S/P Conv.
Removing Cyclic Prefix
Re-mapping Symbols
Cn,0
Data Scatter
Data Gather
Cn,1
Demodulator 2
Cn,N

Demodulator 8
7
Real-rate Implementation (3)
OFDM Demodulator

• MPI Processing

Jedi2
Jedi3
Jedi4
Jedi5
Jedi1
Jedi1
Jedi6
MPI_Send
MPI_Recv
Jedi7
Jedi8
Jedi9
MPI_Recv MPI_Send
8
Real-rate Implementation (4)

• Demo. Environment

OFDM Transmitter (Linux Machine)
OFDM Demodulator 8-node MPI Processing (Linux
Machine)
H.263 Viewer (for DEMO) (Windows Machine)
UDP
TCP
Adaptive Feedback Control
OFDM Modulated Data
H.263 Compressed Video Data
OFDM Modulator (Linux Machine)
9
Real-rate Implementation (5)

• UDP Side
• Packet Size 10,240 Bytes (88(BLSZGUARD))
• OFDM Modulated Sending Data Size 5.14MBytes
• Set UDP receiver side buffer size to 5,500 Bytes
• in order to maximize adaptive control efficiency
• TCP Side
• Packet Size 768 Bytes (8BLSZ3/4)
• H.263 Compressed Stream (after OFDM demodulation)

10
Real-rate Implementation (6)

• Adaptive Feedback Control Algorithm
• Set Initial Sending Period 30ms
• Set initial decrease amount of period to 1ms when
  no error occurs
• (in order to rapidly adapt to the client
  speed during initial stage)
• If there is no error report from the client
  during sending last 10 packets, then it increases
  sending rate
• If there is an error, it increases sending
  interval by the amount proportional to the
  difference between current sending packet number
  and requested packet number, since the difference
  indicates how much the processing speed differs
• and also set decrease amount to 0.5ms after an
  error report
• (in order to adapt accurately)

11
Experimental Environment

• OFDM Transmitter
• marlet.eecom.gatech.edu
• Red-hat Linux V6.2 Kernel 2.4.0
• OFDM Demodulator Using MPI processing
• jedi1jedi9.cc.gatech.edu (9 nodes)
• Red-hat Linux V6.2, Kernel 2.2.17
• LAM 6.3.2 / MPI2 C(University of Notre Dame)
• Demo. Display
• Windows 98 Machine
• Test Data Stream
• ITU-T H.263 Compressed Video Stream
• It runs about 45 seconds

12
Performance Comparison (1)

• Computing Power in MPI Processing

of Packet Retrytimes
Execution Timesec
of MPI machines
13
Performance Comparison (2)
Number of Total Packets 502 Packet Size
10,240 Bytes
Control Policy vs. Transmission Efficiency
14
Performance Comparison (3)
Adaptive Controlled Sending Rate Result (Best
Adaptive)
Sending Period (us)
Packet number
15
Performance Comparison (3)
Adaptive Controlled Sending Rate (start with 3ms
period)
Sending Period (us)
Packet number
16
Summary

• Real-Rate OFDM Receiver Implementation
• Using MPI for High Computing Power
• Adaptive Feedback Control for Data Transmission
• Performance Comparison in MPI Processing
• Investigating Optimal Feedback Algorithm
• H.263 Compressed Data Demonstration