1
Cross-Layer Optimization for
Video Streaming in SingleHop Wireless Networks
Cheng-Hsin Hsu
Joint Work with Mohamed Hefeeda
Simon Fraser University, Canada
MMCN ‘09
January 19, 2009
Video Optimization in Wireless
Networks
2
Resource Allocation Problem
-- shared air medium
Video Optimization Problem
3


Goal: maximize video quality for stations by
properly allocating shared resources
Challenge: stations have diverse constraints
 channel
conditions
 processing powers
 energy levels
 video characteristics

Approach:
 propose
a cross-layer optimization framework
 then, instantiate the framework for 802.11e WLANs
Video Optimization Framework
4
P-R-D Characteristics
Opt. Coding Rate
Complexity
Scalable
Video Coder
APP
The
Optimization
Algorithm
MAC Parameters
Opt. Bandwidth
Share Allocation
Channel Rate
QoS-Enabled
Controller
LINK
Radio
Module
PHY
Complexity Scalable Video Coders
5
Distortion
P-R-D models relate distortion as a func D(.) of rs
(coding rate), ps (coding power),
and V (video characteristics)
Coding Power
Complexity
Scalable
Coder
P-R-D
Characteristics
The
Optimization
Algorithm
Opt.
Coding Rate
Complexity
Scalable
Coder
QoS Enabled Controller
6

Link Layer
 achieves/enforces
QoS differentiation
 allocates bandwidth (bs) to station s, s.t.
where B(.) is the link capacity

b
s
 B,
Physical layer
 diverse
QoS-Enabled
Controller
channel rate (ys)
MAC Parameters
Channel Rate
The
Optimization
Algorithm
Opt. Share of
Bandwidth
QoS-Enabled
Controller
General Formulation
7



Capacity B(.) is a function of
#


Find Opt. policy (rs* ,bs* ) | 1  s  S rs : coding rate
bs : b/w share
such that
S
bs  B(g)
PG : min  s 1 D(g) s.t.


of stations, link protocols, channel rates
Distortion D(.) is a function of
 coding
rate, coding power, video characteristics
General Formulation (cont.)
8

Formulation PG is general
 any
D(.) and B(.) can be adopted
 can be numerically or analytically solved

Different objective functions
 MMSE:
minimizing average mean-square error
PG : min  s 1 D(g)
S

 MMAX:
minimizing maximum distortion
PG : min max Ss1 D()

Instantiate PG for 802.11e WLAN
9


802.11e is a supplement for supporting QoS
Why 802.11e?
 widely
deployed, QoS differentiation, more
challenging than TDMA networks

802.11e supports two modes
 EDCA:
distributed contention-based
 HCCA: polling-based contention-free

Why EDCA?
 simple,
commonly implemented, higher b/w utilization
EDCA Overview
10


QoS differentiation: several Access Categories
(ACs)
Each AC is assigned different back-off parameters
AC
Voice
Video
Best-Effort
Background
AIFS
2
2
5
7
CWmin
3
7
15
15
CWmax
7
15
1023
1023
TXOP
4096
2048
1024
1024
EDCA Overview (cont.)
11



AIFS: Arbitration Inter Frame Space
CW: Contention Window
TXOP: Transmission Opportunity
AIFS
Medium
is Busy
TXOP
Back-off
Time
Transmission
Start
Medium Random Back-off
Time from CW Transmission
is idle
Per-Station QoS Differentiation
12

Assign different EDCA parameters to stations
 CW, AIFS
 TXOP


, then frequency
and bandwidth
, then transmission time
and bandwidth
But, how we choose the EDCA parameters to
achieve a given bandwidth share?
More importantly, how can we estimate overhead
& collisions
Airtime and Efficient Airtime
13


Bandwidth allocation == airtime allocation
Airtime: let  s  rs ys be the fraction of time
allocated to station s
 rs:
application (streaming) rate
 ys: channel rate

Effective Airtime: EA  s the fraction of time
when the shared medium transmits real data
 overhead
and collisions are deducted
Our Effective Airtime Model
14

p-persistent EDCA Analysis [Ge et. Al 07]
 wireless
station draws the back-off time from a
geometric distribution with parameter p
 stateless, so more tractable
2
 analysis can be mapped to EDCA using p 
CW min  2

We develop a closed-form EA model
EDCA
Parameters
Effective
EA Model
Airtime
Effective Airtime Model (cont.)
15
EA 
1
S 1



2S
2  
 1 
1
 
E[x]  CWmin  2  CWmin  


relatively
small
>= 1
where  is a function of several 802.11
parameters, such as AIFS.

Effective Airtime Model (cont.)
16
It is indeed small and can be dropped
802.11e Formulation
17
P : min  s 1 D( ps , rs ,V )
S

s.t.

S
s 1
 s EA
 s  rs ys
where  s  { s  rs ys | 1  s  S}

But, what is D(.)?
MPEG-4 P-R-D Model [He et al. 05’]
18





 s rs p1/3
s
Distortion Ds   2
2
  : video sequence variance
s
  s : coder efficiency
2
s
ps : power consumption
For convenience, we let


s  
 s   s y s p1/s 3
2
s
Optimal Solution
19

Solve it using Lagrangian method for closed-form
solutions
log 2  
*
 S log 2  s s ln 2

 EA
 s1
s




  log 2
s
 s s ln 2
*
s
1
at base station
S
s1
1
s
*
at wireless station
Optimal Allocation Algorithm
20
Base
Station
1,1
*
Wireless
Station 1
1

 2 , 2
*
Compute  * and

adjust TXOP



 3 , 3


Wireless
Station 2
*
Compute  * and
3
adjust TXOP
Wireless
 Station 3
Compute  * and
2
adjust TXOP
OPNET Simulations
21

Implement log-normal path loss model
 more
realistic simulations
 OPNET uses free space model by default

Implement resource allocation algorithms on two
new wireless nodes
 base

station, wireless station
Implement two algorithms
 EDCA (current
algorithm), and OPT (our algorithm)
Simulation Setup
22






deploy 6~8 of wireless stations
wireless stations stream videos to base station
each wireless station periodically (every 5 secs)
reports its status to base station
base station computes the allocation
each wireless station configures its TXOP limit
base station collects stats, such as receiving rate
and video quality
Validation of our EA Model
23
The empirical Effective Airtime
follows our estimation (69%)
Potential Quality Improvement
24
About 28% Distortion Reduction
Dynamic Channel Conditions
25
OPT works in dynamic environments
OPT outperforms opt. algms in ind. layers.
Real 802.11e Testbed
26




Use Atheoros AR5005G 802.11e chip
Implement OPT and EDCA algorithms in its Linux
driver
Configure a base station and two wireless stations
Wireless stations report status every 10 sec
Sample Result from the Testbed
27
Quality improvement:
up to 100% in MSE, or 3 dB in PSNR
Conclusions
28





Proposed a general video optimization framework
Instantiated the problem for 802.11e networks
Presented models for 802.11e and developed an
effective airtime model
Analytically solved the optimization problem
Evaluated the solution using OPNET simulator and
real implementation. Both show promising quality
improvements.