Feedbackbased Adaptive Video Transmission over wireless network

1
Feedback-based Adaptive Video Transmission over
(wireless) network
Presentation in CS 6235 (Real-Time Systems)
April 2001 Wanxia Xie College of Computing
2
Outline

• Introduction
• Overview
• Architecture w/o adaptation
• Architecture w/ adaptation
• Implementation
• Monitor (Feedback Controller) / UDP
  communication / Monitor Channel /Adaptation
• Results
• Future Work
• Conclusion

3
Introduction

• Problem with Video transmission over (wireless)
  network
• Dynamic traffic intensity and bandwidth (Low
  bandwidths, high error rates and disconnection
  for wireless network)
• Dynamic system resources (cpu, memory, power)

4
Old Architecture w/o adaptation
Control Channel
Raw Channel
Mpeg Channel
Gen
Play
Trans
Socket (TCP)
Xpalm
5
New Architecture w/ adaptation
Control Channel
Raw Channel
Mpeg Channel
Xpalm w/probe
Play w/ Probe
Gen
Trans
Monitor (Feedback Controller)
Socket (UDP)
Monitor Channel
6
Probe and Feedback Collection

• Mpeg Queue Fill Level in play
• CPU usage and memory usage in play
• Raw Queue Fill Level in xpalm
• Packet Loss Rate in xpalm
• CPU usage and memory usage in xpalm

7
Adaptation

• Adaptation for play
• 1)Downgrade Quality of Raw Video
• 2)Compress / Decompress of Raw Video (If CPU
  usage and memory usage in play and xpalm is low)
• 3)Change raw video transmission rate (Bottleneck
  is decode process from mpeg to raw video)
• Adaptation for gen
• Change mpeg transmission rate in gen

8
Monitor (Feedback Controller)

• Receive feedback information from monitor channel
• Make the adaptation decision and change control
  parameters such as quality criteria,
  compress/decompress option and transmission rate
• Send the updated control information to control
  channel.
• (Receiver of Control information, gen and play
  update their control information and apply these
  in mpeg or/and raw video transmission)

9
Experiment Setup
Red Hat Linux 6.2 (kernel 2.2.13) on an
i686. trans runs in edhpc1.cc.gatech.edu xpalm
runs in edhpc8.cc.gatech.edu play runs in
edhpc3.cc.gatech.edu gen runs in
edhpc4.cc.gatech.edu monitor runs in
edhpc5.cc.gatech.edu ( CPU and memory sampling
every second) Experiment Goal (Solve the
bottleneck) 1) Maintain mpeg queue fill level
60-80 2) Adapt video transmission for dynamic
network activity
10
Result (1) Mpeg transmission rate w/o adaptation
11
Result (2) Mpeg Queue fill level w/o adaptation
12
Result (3) Mpeg Packet Loss Rate w/o adaptation
13
Result (4) Mpeg transmission rate w/ adaptation
(buf 1000)
14
Result (5) (No packet loss)Queue fill level w/
adaptation (buf 1000)
15
Result (6) Mpeg transmission rate w/ adaptation
(buf 100)
16
Result (7) Queue fill level w/ adaptation (buf
100)
17
Result (8) Mpeg Packet Loss Rate w/
adaptation(buf 100)
18
More results

• Mpeg transmission rate, queue fill rate, mpeg
  packet loss rate for buf 500.
• Latency of Monitor (feedback controller)
• CPU usage and memory usage with adaptation
  Compress/decompress (cpu usage 100)
• Quality adpatation for xpalm
• Packet loss rate in xpalm (UDP connection) (For
  High contention, middle contention and minimum
  contention)

19
Future Work

• Multiple video streams
• PID control algorithm
• Dynamic sampling frequency for probe to reduce
  the overhead of probe
• Real wireless network simulation
• Good sample application for infopipe
• Generic interface for infopipe specification

20
Conclusion

• Maintain the queue buffer fill level
• Reduce packet loss rate
• Flexible adaptation (quality adaptation,
  compress/decompress, transmission rate)

21
Reference

• 1)