Transcript slides

On the (In)feasibility of Fine
Grained Power Control
Vivek Vishal Shrivastava
Dheeraj Agrawal
Arunesh Mishra
Suman Banerjee
Tamer Nadeem (Siemens Research)
Department Of Computer Sciences
University of Wisconsin-Madison
Sep 27, 2006
MobiCom SRC ‘06
Transmission Power Control
Spectral
Efficiency
Energy
Efficiency
High Power
Low Power
Sep 27, 2006
MobiCom SRC ‘06
Transmission Power Control
 A wide variety of power control algorithms have been
proposed in literature
 Few have made it to practice
 This gap has been attributed to lack of sophisticated
hardware
 Absence of fine grained power levels in current state of
the art wireless cards
Sep 27, 2006
MobiCom SRC ‘06
Transmission Power Control
 A wide variety of power control algorithms have been
proposed in literature
 Few have made it to practice
 This gap has been attributed to lack of sophisticated
hardware
 Absence of fine grained power levels in current state of
the art wireless cards
Our claim: Even if fine-grained power control was
available in wireless cards, no algorithm will be able
to take advantage of it in any practical setting due to
significant RSS variations
Sep 27, 2006
MobiCom SRC ‘06
The Essence
Q. What granularity of power control is
practically usable and how do we determine
these discrete power levels ?
Sep 27, 2006
MobiCom SRC ‘06
The Essence
Q. What granularity of power control is
practically usable and how do we determine
these discrete power levels ?
A1. In practical settings, significant overlap
between RSS for different power levels
makes fine grained power control infeasible
Sep 27, 2006
MobiCom SRC ‘06
The Essence
Q. What granularity of power control is
practically usable and how do we determine
these discrete power levels ?
A1. In practical settings, significant overlap between
RSS for different power levels makes fine grained
power control infeasible
A2. Few carefully chosen, environment
dependent, discrete power levels are
practically usable
Sep 27, 2006
MobiCom SRC ‘06
In this talk, we substantiate these
claims and build an empirical power
control model on the basis of these
guidelines
Sep 27, 2006
MobiCom SRC ‘06
Some Existing Power Control Approaches
PCMA [Infocom ‘01]
• An interesting work that proposed use of power control for
throughput enhancement
• Designed power controlled medium access
• Receiver finds optimum power and sends a feedback to
the transmitter
• Use of out-of-band busy tones to silence neighbors
Other approaches: SHUSH[WICON ‘05], IPMA[SCC 2003]
Sep 27, 2006
MobiCom SRC ‘06
Some Existing Power Control Approaches
PCMA [Infocom ‘01]
• One of the first works to use power control for throughput
enhancement
• Designed power controlled medium access
• Receiver finds optimum power and sends a feedback to
the transmitter
• Use of out-of-band busy tones to silence neighbors
Other approaches: SHUSH[WICON ‘05], IPMA[SCC 2003]
Works well with fine grained power control
What happens if RSS variations are present?
Sep 27, 2006
MobiCom SRC ‘06
Limitations
Use of fine grained power levels works
well in the absence of RSS variations
Sep 27, 2006
MobiCom SRC ‘06
Limitations
Use of fine grained power levels works
well in the absence of RSS variations
However, RSS variations are significant
in typical wireless scenarios
Sep 27, 2006
MobiCom SRC ‘06
RSS Variations
Multipath, fading,
shadowing
Sep 27, 2006
External
Interference
MobiCom SRC ‘06
RSS Variations
Multipath, fading,
shadowing
overlap
Sep 27, 2006
External
Interference
20% packets are
received at RSS of
22dBm
MobiCom SRC ‘06
RSS Variations
Multipath, fading,
shadowing
Sep 27, 2006
External
Interference
MobiCom SRC ‘06
40,50,60 mw
have
significant
overlap
RSS Variations
Multipath, fading,
shadowing
External
Interference
with
without
Sep 27, 2006
MobiCom SRC ‘06
Implications of RSS variations
• Receiver cannot distinguish two transmit power levels with
significant overlap
• Only transmit power levels with minimum overlap be used
together
• Needs some number of packets (>1) to characterize RSS
distribution
Sep 27, 2006
MobiCom SRC ‘06
The Essence - Part I
Q. What granularity of power control is
practically usable and how do we determine
these discrete power levels ?
A1. In practical settings, significant overlap
between RSS for different power levels makes
fine grained power control infeasible
Sep 27, 2006
MobiCom SRC ‘06
RSS variations are environment
dependent
Line of
Sight
Non Line
of Sight
Non Line of Sight
with Hotspot
Interference
Non Line of Sight
with controlled
interference
Sep 27, 2006
MobiCom SRC ‘06
Practical Transmit Power Control
Sample sufficient
number of packets
at each power level
Sep 27, 2006
MobiCom SRC ‘06
Practical Transmit Power Control
Sample sufficient
number of packets
at each power level
Sep 27, 2006
Characterize
RSS
distribution
MobiCom SRC ‘06
Practical Transmit Power Control
Sample sufficient
number of packets
at each power level
Sep 27, 2006
Characterize
RSS
distribution
MobiCom SRC ‘06
Operate on power
levels with nonoverlapping RSS
distributions
Characterizing RSS distribution
What is the minimum sample size to accurately
capture RSS distribution?
Sep 27, 2006
MobiCom SRC ‘06
Characterizing RSS distribution
What is the minimum sample size to accurately
capture RSS distribution?
– RSS variations are typical of a particular
indoor environment
– Different number of packets may be required
to accurately capture RSS distribution
– Brute Force : Capture very large number of
packets for determining RSS distribution
Sep 27, 2006
MobiCom SRC ‘06
Characterizing RSS distribution
What is the minimum sample size to accurately
capture RSS distribution?
– RSS variations are typical of a particular
indoor environment
– Different number of packets may be required
to accurately capture RSS distribution
– Brute Force : Capture very large number of
packets for determining RSS distribution
Can we do better ?
Sep 27, 2006
MobiCom SRC ‘06
Online Mechanism
Normalized Kullback-Leibler Divergence (NKLD)
Quantifies the distance or relative entropy between two
distributions
Operating point
Sep 27, 2006
MobiCom SRC ‘06
Online Mechanism
Sample n + δ packets
If divergence < threshold
return the distribution
Sep 27, 2006
Calculate distribution
of RSS for n, n + δ
Compute divergence using
statistical tools like NKLD
MobiCom SRC ‘06
Evaluation accuracy of RSS distributions obtained
with Online Mechanism
Sep 27, 2006
MobiCom SRC ‘06
Online Mechanism
 Sample sufficient number of packets, to
capture RSS distribution with some accuracy
 Profile different available power levels
 Find the power levels with non overlapping
RSS distribution
 Repeat this procedure periodically to cope
up with large scale variations in channel
conditions
Sep 27, 2006
MobiCom SRC ‘06
Experimental Testbed
12
10
8
11
Sep 27, 2006
MobiCom SRC ‘06
The final outcome
Number
of power
levels
3
1
2
3
Feasible Power Levels at four receivers in the testbed
Sep 27, 2006
MobiCom SRC ‘06
The Essence – Part II
Q. What granularity of power control is
practically usable and how do we determine
these discrete power levels ?
A1. In practical settings, significant overlap between
RSS for different power levels makes fine grained
power control infeasible
A2. Few carefully chosen, environment
dependent, discrete power levels are
practically usable
Sep 27, 2006
MobiCom SRC ‘06
Future Work
 Use our model as a module in previously
proposed Transmit Power Control
mechanisms
Sep 27, 2006
MobiCom SRC ‘06
Future Work
 Use our model as a module in previously
proposed Transmit Power Control
mechanisms
 Study the interdependence between
power and data rates, in view of few
discrete power levels
Sep 27, 2006
MobiCom SRC ‘06
Future Work
 Use our model as a module Transmit
Power Control mechanisms
 Study interdependence between power
and data rates, in view of few discrete
power levels
 Build a practical transmit power
control mechanism using the
guidelines discussed here
Sep 27, 2006
MobiCom SRC ‘06
Questions ?
Sep 27, 2006
MobiCom SRC ‘06