Energy efficient algorithms and techniques for wireless mobile clients 0

POWER MANAGEMENT AT NETWORK INTERFACE LEVEL... In modern smartphones, the three main sources of power consumption are, 1 the display, 2 the network interfaces, and 3 the CPU.. We found t

ENERGY EFFICIENT ALGORITHMS AND

TECHNIQUES FOR WIRELESS MOBILE CLIENTS

BHOJAN ANAND

(Ph.D.), NUS

A THESIS SUBMITTED FOR THE DEGREE OF DOCTOR OF PHILOSOPHY

DEPARTMENT OF COMPUTER SCIENCE

SCHOOL OF COMPUTINGNATIONAL UNIVERSITY OF SINGAPORE

2012

I hereby declare that the thesis is my original work and it has been written by me in

its entirety I have duly acknowledged all the sources of information which have

been used in the thesis

This thesis has also not been submitted for any degree in any university previously

———————————–

Bhojan Anand

8 August 2012

iii

iv

First of all I would like to thank my supervisor Professor A.L.Ananda I appreciate his guidance and support not only my research butalso my life After four years of his supervision, his impressive

leadership becomes a big milestone in my life

I am indebted to the coauthors of several papers included in this thesis.The coauthors are Associate Professor Chan Mun Choon, AssociateProfessor Rajesh Krishna Balan, Associate Professor Ooi Wei Tsang,Associate Professor Chang Ee Chien, Mr Pravein Govindan Kannan,

Mr Karthik Thirugnanam, Mrs Jeena Sebastien, Mr Le Thanh Long, MrPham Duc-Dung, Mr Soh Yu Ming and Mr Chong Lee Kee

A special thanks to Associate Professor Ooi Wei Tsang and AssociateProfessor Roger Zimmerman for their productive comments for my thesisproposal and research I am also particularly grateful to AssociateProfessor Chan Mun Choon for his constructive comments and support

throughout the research period

I appreciate to all fellow Ph.D students and friends at the

Communication and Internet Research Lab for making the supportivework environment My deepest gratitude goes to my wife Florence Anandand sons Mikhil Anand and Sashil Anand for their unflagging love andsupport throughout my life; this dissertation is simply impossible without

them

v

TABLE OF CONTENTS

Page

DECLARATION iii

ACKNOWLEDGEMENTS v

SUMMARY xii

LIST OF TABLES xv

LIST OF FIGURES xvi

LIST OF PUBLICATIONS xxix

CHAPTER 1 INTRODUCTION 1

1.1 Mobile Clients Space 1

1.2 Mobile Applications 3

1.3 Saving Energy 4

1.3.1 Why is saving energy in mobile clients important? 6

1.3.2 Current Status & Challenges 9

1.3.2.1 Display Power Conservation 10

1.3.2.2 Wireless Interface Power Conservation 12

1.4 Thesis Contribution 13

1.5 Thesis Organisation 16

2 RELATED WORK 18

2.1 LCD Power Conservation 18

2.2 OLED Display Power Conservation 33

2.3 Network Interface Power Conservation 43

2.4 Processing Unit Power Conservation 61

2.5 Summary 63

vi

3 DISPLAY POWER MANAGEMENT (LCD) 65

3.1 LCD Display Technology 65

3.2 Tone Mapping Technique & Its Advantages 68

3.3 Using the Gamma Function for Tone Mapping 70

3.4 System Design 73

3.4.1 Backlight Power Measurement 75

3.4.2 Gamma to Backlight Relationship 76

3.4.2.1 Analysis of Gamma to Backlight Relationship 77

3.4.3 Measuring Image Quality 78

3.4.3.1 Using the Image Quality Metrics 81

3.4.4 Computing Image Brightness 82

3.4.5 Human Calibration of Gamma Thresholds 83

3.4.5.1 Objective 83

3.4.5.2 Methodology and Setup 83

3.4.5.3 Results 85

3.4.6 Objective Analysis of Gamma Thresholds 86

3.4.7 Run-time Algorithm 87

3.5 Implementation 88

3.5.1 Selection of Games 88

3.5.2 Changing Backlight Level and Gamma 90

3.6 Evaluation Methodology 90

3.6.1 Power Measurement Testbed Setup 91

3.6.2 Power Measurement - Methodology 93

3.6.3 User Study - Methodology, Participants, and Setup 94

3.7 Evaluation Results 96

3.7.1 Baseline Measurements 96

3.7.2 Measured Analytical Results 99

3.7.3 User Study Results 100

3.7.4 Overall Result: System Works Very Well 100

3.7.5 Summary 102

vii

4 DISPLAY POWER MANAGEMENT (OLED) 103

4.1 OLED Display Technology 104

4.2 Key Observations on OLED Displays 105

4.3 Power Optimisation for Webpages - Texts 110

4.3.1 Colour Harmony 111

4.3.2 Brand Colour & Brand Identity 113

4.3.3 Chromatic and Achromatic Contrast & Colour Mapping 120

4.4 Power Optimisation for Webpages - Images 124

4.4.1 Luminance Adaptive Colour Transformation 126

4.4.2 HVS based Colour Transformation Algorithm 131

4.4.2.1 Algorithm Alternative 134

4.4.3 Adapting to Other Contents 138

4.5 System Implementation 139

4.6 Evaluation Methodology 141

4.6.1 Quality Measurements - Objective Metrics 142

4.6.2 Quality Measurements - Subjective User Study 142

4.7 Evaluation Results 145

4.7.1 Evaluation Results of Colour Transformed Webpages 145

4.7.2 Evaluation results of HVS based Image Manipulation Algorithm 146

4.7.3 Overall Result (Combined) 150

4.7.4 Summary 151

5 POWER MANAGEMENT AT NETWORK INTERFACE LEVEL 152

5.1 Mobile Games and Game Maps 153

5.1.1 Binary Space Partitioning 154

5.1.1.1 Potentially Visible Set 155

5.1.1.2 Limitations 158

5.1.2 Quadtree and Octtree 158

5.2 Distance Based Approach 161

viii

5.2.1 Game State Estimation 161

5.2.2 Macro Power Management 162

5.2.3 Micro Power Management 166

5.3 Visibility Based Approach 169

5.3.1 Dynamic Lookahead 172

5.3.2 Determining Sleep Times and Intervals 175

5.4 3D Renderer’s View Based Approach 176

5.4.1 Visibility and Spatial Subdivision Scheme 177

5.4.2 Two-Level Scan Algorithms 178

5.4.2.1 Macro Scanning Algorithm 179

5.4.2.2 Micro Scanning Algorithm 182

5.5 Wireless Interface Control at Client Side 183

5.6 Algorithm Selection 184

5.7 Implementation 185

5.7.1 Network Characteristics of FPS and MMOG Games 185

5.7.2 Sleep Command 187

5.8 Evaluation Methodology 188

5.8.1 Evaluation Objectives 191

5.8.2 Defining a Quality & Power Metric 192

5.8.3 Small Scale User Study - Methodology 195

5.8.4 Experiments 197

5.8.5 Additional Evaluations - Visibility Based Approach 198

5.8.5.1 Using Traces for Repeatability 198

5.8.5.2 Running the Simulations 199

5.8.6 Additional Evaluations - Renderer’s View Based Approach 199

5.8.6.1 Using AMID for Error Control 200

5.9 Evaluation Results 203

5.9.1 Distance Based Approach 203

5.9.1.1 Sparse Environment - Low player Density 204

5.9.1.2 Dense Environment - High Player Density 205

ix

5.9.1.3 Contribution from Micro and Macro Power

Management 205

5.9.2 Visibility Based Approach 205

5.9.2.1 Baseline - No Prediction 206

5.9.2.2 Experiments over Various Networks 208

5.9.2.3 Effect of Velocity 210

5.9.2.4 Effect of Different Sleep/Wakeup Intervals 212

5.9.2.5 Effect of player density 213

5.9.2.6 Benefit of Our Dynamic Algorithm 214

5.9.2.7 Real Power Measurements 215

5.9.2.8 Another Perspective of Power Savings 216

5.9.2.9 Impact of Errors on Perceived Quality 217

5.9.3 3D Renderer’s View Based Approach 218

5.9.3.1 Effects of Map Type 219

5.9.3.2 Effects of Energy Threshold 220

5.9.3.3 Effects of Player Density (number of Players) 221

5.9.3.4 Effects Error Controller on Optimising Algorithm Parameters 222

5.9.3.5 Effects Error Controller on Average Error Rate 224

5.10 Summary of all Results 226

6 CONCLUSIONS AND FUTURE WORK 227

6.1 Conclusions 227

6.1.1 Conserving LCD Energy 228

6.1.2 Conserving OLED Display Energy 229

6.1.3 Conserving Wireless Network Energy 229

6.2 Directions for the Future Research 230

BIBLIOGRAPHY 235

APPENDICES A GAMMA CALIBRATION (FOR LCD) - SURVEY FORM 259

B DISPLAY POWER MANAGEMENT (LCD) - SURVEY FORM 265

C DISPLAY POWER MANAGEMENT (OLED) - SURVEY APPLICATION AND FORM 273

x

D NETWORK INTERFACE POWER MANAGEMENT

-SURVEY FORM 280

xi

ENERGY EFFICIENT ALGORITHMS AND

TECHNIQUES FOR WIRELESS MOBILE CLIENTS

BHOJAN ANAND

(Ph.D.), NATIONAL UNIVERSITY OF SINGAPORE

Directed by: Professor Dr A.L Ananda

In this thesis we present a suite of algorithms and techniques for conserving energy

in battery operated smartphones In modern smartphones, the three main sources of

power consumption are, 1) the display, 2) the network interfaces, and 3) the CPU

We found that the wireless interface and display components dominate the power

consumption in current generation of smartphones

Firstly, we focus on an adaptive backlight controlling algorithm which uses tone

mapping techniques intelligently for managing LCD display power of these devices

The uniqueness of our work is the use of an efficient tone mapping operator γ (Gamma

Correction) to brighten the content γ is implemented in hardware level for modern

smartphones and hence, it is efficient However, due to its non-linearity, we conducted

set of well controlled experiments to get the relationship between γ and brightness

In addition, we found that the contents which are brighter tend to loose more global

xii

contrast than darker contents when same amount of γ is applied The effect of contrast

loss fostered us to develop an algorithm for dynamic application of gamma controlled

by contrast loss that can save up to 68% of energy

Secondly, we describe colour transformation service, which transforms the colours

of web contents to power efficient colours for OLED displays We found that power

consumption of OLED displays in smartphones depends on the colour and intensity of

the contents We used the principle of colour theory (in particular, colour

harmonic-ity) and the characteristics of human visual system (in particular, eye’s non-liner

sensitivity to visible light spectrum and luminance levels) for colour transformation

and found that we can generate webpages with good colours (sometimes, even better

than the original colours of the webpage) that are power efficient and pleasing for

eyes We show that with our approach we can save more than 60% of energy while

ensuring pleasant reading experience

Thirdly, we discuss our algorithms which use three different approaches to manage

power at network interface level and a scheme for selecting the algorithms based on

the game genre and game map The three approaches are, distance based approach,

visibility based approach and 3D renderer’s view based approach The key goal of

all these algorithms is to predict the importance of client’s game state for next t

milliseconds If a client has no chance for interacting with other player’s for next t

milliseconds the wireless interface is put into low power mode We found that distance

based algorithm is efficient for huge open maps with few occlusions and the other two

are good for highly occluded maps It is interesting to find that games with highly

occluded maps such as first person shooting games give more chances for sleeping We

xiii

can save about 57% of wireless interface energy with highly occluded maps without

affecting the quality of game play adversely Renderer view based approach gives the

most efficient result provided the map is highly occluded and supportive occlusion

culling algorithm is already a part of the game engine

xiv

LIST OF TABLES

1.1 Battery Lifetime in Modern Smartphones 5

2.1 Power consumption (in Watts) for various modes of Intel (3x3 MIMO) and Atheros (2x2 MIMO) NICs 52

3.1 Demographics Statistics for the User Study 97

3.2 Baseline Power Consumption of the Laptop 98

3.3 Baseline Power Consumption of the HTC Magic 98

3.4 Power-Savings Measurements 99

4.1 Demographics Statistics for the User Study 143

5.1 Map Size of MMOG games 158

5.2 Power Characteristics of Different Interfaces 184

5.3 Selection of Algorithms 186

5.4 Power characteristic of the card used 189

5.5 Effects of Map Type - Experiment Variable Setup 219

xv

LIST OF FIGURES

1.1 Backlight level vs Power for HTC Magic Android Phone [1] 6

1.2 Component Power Consumption (HTC Magic) [1] 9

2.1 Structure of a Transmissive TFT LCD 19

2.2 Visibility of the Image in a Transmissive TFT in some Environment Luminance Condition [2] 20

2.3 Image and its Discrete Histogram 21

2.4 Luminance as a function of Backlight and Transmissivity 24

2.5 Visual Effects of Adjusting Brightness (b), Contrast (c), and Both (d) when the Backlight is Dimmed to 50% 25

2.6 Luminance as a function of Backlight and Transmissivity [3] 27

2.7 Luminance vs Perceived Brightness 28

2.8 Relation Between MSE and Backlight Level 32

2.9 Energy plot of the hues of 6 categorical colours (from left to right: blue, red, purple, orange, green, yellow) Evergy (E) vs Lightness (L∗) [4] 35

2.10 Categorical colours of varying lightness sorted by increasing energy cost [4] 35

2.11 Tooth Dataset Coloured with Traditional Colours (Original) and Energy Efficient Colours [4] 36

2.12 Unstructured Transformed GUIs with Different Settings 38

2.13 Colour Transformed ESPN Webpage 39

xvi

2.14 Colour Quantizing an Image to N = 32 Colours 40

2.15 Colour Quantizing an Image to N = 512 Colours 42

2.16 Power Consumption of Images Generated using Different Algorithms 43

2.17 Power-constrained LMHE 44

2.18 802.11 State Transition 46

2.19 Client Listening Intervals in BSD 48

2.20 Drop Rate vs Energy Metric (Real player format at 512 Kbps) [5] 51

2.21 Split Communication for Multimedia Streaming 53

2.22 Cell-to-Notify Protocol 56

2.23 Suspension Period 60

2.24 Suspension Period 61

3.1 Transmissive LCD Displays 66

3.2 Reflective LCD Displays 67

3.3 The Effect of Gamma and Linear Transformations The Amount of Power Saved is the Same for Both Approaches 71

3.4 Global Contrast Loss vs Gamma 73

3.5 Global Contrast Change vs Image Brightness for γ = 3 74

3.6 Power vs Backlight level 75

3.7 Backlight level vs Gamma value 77

3.8 Effect of Gamma Increase and Compensation 78

3.9 Gamma and Backlight Adjusting Tool 84

3.10 User-Perceived Acceptable Gamma Levels 86

3.11 Contrast Loss vs Adaptive Gamma 87

xvii

3.12 Flowchart of our System 89

3.13 Testbed 92

3.14 Setup for Laptop Power Measurement (Overall) [Note: Lenovo Thinkpad W500 Laptop Adapter outputs 20V DC] 92

3.15 Results of the User Study for all 3 Maps All Versions of the Game were Deemed Playable by the Participants 101

4.1 OLED Energy Consumption vs Screen Brightness 106

4.2 OLED Energy Consumption vs Gamma Value 107

4.3 AMOLED sub-pixels close-up 108

4.4 Energy Vs RGB Sub-Pixel Values 110

4.5 Colour Wheel in RGB Colour Space 112

4.6 Colours Wheel Types 114

4.7 Colours - Associated Words - Sample Logos 115

4.8 Webpages designed using Brand Colours available in their Logos 116

4.9 Logos are Used as Favicons 118

4.10 Colour Extraction and Ranking from NUS favicon 119

4.11 Colour Extraction and Ranking from INTEL favicon 119

4.12 Background vs Text Area (After Excluding Images) - Sample Webpages 123

4.13 Background vs Text Area 124

4.14 Change in Contrast vs Gamma 126

4.15 Effects of basic approaches on Image Contrast 127

4.16 Pixel power models of Goole Nexus One, Samsung Galaxy S and Nokia N85 OLED displays Axis X represents gamma-corrected linear RGB values [6] 129

xviii