Optimal Fixed

1
Optimal Fixed Scalable Energy Management for
Wireless Networks

• Rahul Mangharam
• rahul_at_cmu.edu
• S. Pollin, B. Bougard, R. Rajkumar,
• F. Catthoor, L. Van der Perre, I. Moerman
• Carnegie Mellon University, USA IMEC, Belgium
• INFOCOM - March 15, 2005

2
The Device Energy Crisis
Source 1

• Processor Power Consumption 200/4 years
• Battery Energy Density 25/
  4years

1 K. Lahiri, A. Raghunathan, S. Dey, D.
Panigrahi "Battery-Driven System Design A
New Frontier in Low Power Design", VLSI Design,
Jan. 2002
3
The Wireless Transceiver Energy Crisis
x 4.0hrs 4.0 Whr x 2.0hrs 2.6 Whr ?7.5Whr x
0.5hrs 0.9 Whr

• How long will a battery last today?
• Handhelds weight lt 350g. (12oz)
• Batteries
• Battery Weight lt 50 of handheld (175g)
• Capacity 90Whr/Kg (NiCd) ? 15Whr
• 802.11a Transceiver consumes 7.5Whr or 50
• Transceivers consume upwards of 30 of a
  laptops overall energy

4
Outline

• The Cross-Layer Energy Management Problem
• MEERA Methodology for Energy Efficient Resource
  Allocation
• Theoretical Foundations
• Two-phase Solution Approach
• Design Time
• Run-Time
• System Case Study 802.11a Transceiver
• Energy Vs Performance Results

5
Focus on MAC PHY
Application
Transport
Network
MAC Scheduling
PHY Comms
PHY - RFIC
PHY Channel
6
The Cross-Layer Energy Management Problem

• Leverage ALL Control Knobs
• Real-Time MPEG-4 Stream
• Sleep-Aware MAC
• Modulation, Code Rate, Pkt Sz, etc
• Tx Power, PA Back-off
• 5-state Channel Model

SYSTEM STATE
SCALING
SLEEPING
7
The Cross-Layer Energy Management Problem

• What system configurations to assign to each user
  at runtime to minimize energy consumption while
  providing a sufficient level of QoS?
• Given a shared fading channel and multiple users
  with bursty delay-sensitive data

Data transmission Channel access grant
AP
Uplink
Node 1
Node 4
Peer-to-peer link
Node 2
Node 3
8
To SLEEP or To SCALE?
No Retry 1 Retry 2 Retry
Tx Power
Idle Power
E n e r g y
Time Slot
9
MEERA Methodology for Energy-Efficient
Resource Allocation

• Design Time
• Profile Energy vs Control Dimensions (K)
• Profile Time vs Control Dimensions
• ?Obtain Energy-Time Trade-off
• Prune Non-Convex Operating Points

For Each Node
10
MEERA Optimal Cross Layer Sleep-Scaling

• Proved Bounded Small Deviation from Optimal
  2
• Simple Greedy algorithm
• Following segments with steepest slope

2 R. Rajkumar, C. Lee, J. Lehoczky and D.
Siewiorek "A Resource Allocation Model for QoS
Management" IEEE RTSS, 1997.
11
Greedy Runtime Algorithm
Access Point
Data transmission Channel access grant
Node 1
Node 4
Node 3
Node 2
12
Greedy Runtime Algorithm - Operation
At Base Station
Power
Period 30 ms
13
Case Study 802.11a Transceiver

• Metrics
• Job Failure Rate (JFR)
• Packet Error Rate (PER)
• Bit Error Rate (BER)
• Symbol Error Rate
• SINAD SNR with Added Distortion

• For a given Target JFR and Channel State
• Determine the Control Dimensions
• with Lowest Energy Consumption

14
PHY RF Circuit Modeling
SINAD
Efficiency
Profile of Microsemi LX5506 802.11a HPA
15
PHY Channel Model
BER
SNR Channel State
5-state Indoor Channel Model derived from actual
experiments
16
MEERA Cost ?? Resource Trade-off
Energy
Time
TX Power
TX Power
PA Back-off
PA Back-off
The mapping for the PA output power and back-off
control dimension for a fixed setting of the
modulation and code rate control dimensions.
17
Energy-Aware MAC
Data transmission Channel access grant
AP
Uplink
Node 1
Node 4
Peer-to-peer link
Node 2
Node 3















Time

18
Performance Impact of System State
Large Energy Range within and across system
states
19
Performance Impact of Link Utilization

• As flows increase, each flow uses a smaller
  TXOP
• MPEG consumes more energy than CBR for same
  average rate

20
The Need for Sleep/Scaling

• To meet higher SINAD Requirement
• Lower Modulation
• Higher Transmit Power
• Increased cost of retransmissions

21
Performance Time-varying Channel

• Varying channel consumes more energy than static
  channel
• Energy savings split evenly between Sleeping and
  Scaling

22
Conclusion

• MEERA A General Methodology for Energy-Efficient
  Cross Layer Design
• Control Dimensions ? Cost (Energy)
• Control Dimensions ? Resource (Time)
• Design Time Cost-Resource Profiles
• Run-Time Greedy Resource Allocation
• Bounded Optimal Results
• 802.11a Transceiver Case Study
• Actual Channel
• Real RF Circuits
• Communications Layer
• Energy-aware MAC
• Real-Time Application
• 2-9X Savings in Energy Sleep Scale!

23
MEERA Formal Problem Statement
Assumption all flows i can be scheduled under
worst-case conditions
24
Backup Slides
25
Case Study System Overview
26
MEERA Definitions

• System consists of n flows Fi, .., Fn (i1,..,n)
  described with
• Cost function Ci
• gt to minimize
• QoS Constraint Qi
• gt to meet
• Shared Resource(s) Ri,j j1,..,m
• gt to share
• Control Dimensions Ki,r r 1,..,l
• gt to set
• System States Si,t t 1,..,s
• gt to take
  into account