Advanced control technologies for future storage devices

28 3 Servo Control Design for a High TPI Servo Track Writer with Microactuators 30 3.1 Introduction.. 49 4 A Unified Control Scheme for Combined Seeking and track follow-ing of a HDD Ser

ADVANCED CONTROL TECHNIQUES FOR

FUTURE STORAGE DEVICES

THUM CHIN KWAN

NATIONAL UNIVERSITY OF SINGAPORE

2008

THUM CHIN KWAN

(B.Eng (Hons.), NUS )

A THESIS SUBMITTEDFOR THE DEGREE OF DOCTOR OF PHILOSOPHYDEPARTMENT OF ELECTRICAL AND COMPUTER

ENGINEERINGNATIONAL UNIVERSITY OF SINGAPORE

2008

my parents, my sisters, my brothers-in-law,

and my friends

I would like to express my gratitude and respect to my advisor, Professor Ben

M Chen for his guidance and support His deep insights and broad views oftengive me new and valuable ideas in my research If not for him, the successfulcompletion of this research would be virtually impossible I would also like to thank

Dr Du Chunling of A*STAR Data Storage Institute (DSI) for her sincere and timelyassistance in addition to her beneficial instructions during the course of this research.Her comments and suggestions have been extremely useful to me for improving thequality of my research Besides them, I am also deeply indebted to my co-advisor,

Dr Ong Eng Hong of DSI as well as my former co-advisor, Dr Guo Guoxiao ofWestern Digital Technology for their guidances into the field of Hard Disk Drive(HDD) technologies

I am very grateful to many DSI staffs and students, namely, Dr Teoh Jul Nee,

Mr Lai Chow Yin as well as former DSI students, Dr Pang Chee Khiang and

Dr Zheng Jinchuan for how much I had benefited from so many of our discussions

on HDD servo technologies I am also grateful to Dr Peng Kemao and Dr LumKai-Yew of Temasek Laboratories, National University of Singapore (NUS), for theirvaluable suggestions, which aided in the development of some ideas in this research

To my family, I want to thank them for granting me the absolute freedom

to choose the career path I desire To my friends, I wish to express my deepestappreciation for their unconditional and relentless encouragement and emotional

i

ACKNOWLEDGEMENTS ii

support throughout the course of this research, especially during those difficult times

at work and love

Finally, I want to thank NUS for offering me the NUS Research Scholarshipfor my Ph.D study in NUS, and DSI for providing the experimental platform andequipment for my research as well as those two competition trophies that help todecorate my cubicle quite nicely

Acknowledgements i

1.1 Background and Motivation 1

1.2 Hard Disk Drives 3

1.3 HDD Servo Systems 6

1.3.1 Servo Information 6

iii

CONTENTS iv

1.3.2 Servo Control 7

1.4 Servo Track Writing Technology 14

1.5 Contributions of the Thesis 14

1.6 Organization of the Thesis 17

2 System Identification and Implementation Setup 18 2.1 HDD Implementation Setup 18

2.2 Plant Modeling 20

2.2.1 HDD VCM Actuator Modeling 21

2.2.2 STW Platform 23

2.3 Noise and Disturbance of STW platform 27

2.4 Conclusion 28

3 Servo Control Design for a High TPI Servo Track Writer with Microactuators 30 3.1 Introduction 30

3.2 Generalized KYP Lemma and Its Original Application in Sensitivity Function Shaping 33

3.3 An Enhanced Generalized KYP Lemma Based Sensitivity Function Shaping Technique 35

3.3.1 Selection of Poles of Q(z) and Simulation Results 36

3.3.2 H∞ Stability Robustness 42

3.4 Rejecting Low-Frequency Disturbances for a STW Platform: Design and Implementation Results 44

3.5 Conclusion 49

4 A Unified Control Scheme for Combined Seeking and track follow-ing of a HDD Servo System 50 4.1 Introduction 50

4.2 Unified Control Scheme for Track Seeking and Track Following and Controller Design 53

4.2.1 Controller Structure 55

4.2.2 Stability Issues 57

4.2.3 Design of Controller Parameters for HDD Servo System 62

4.2.4 Controller Design for HDD Application 65

4.3 Simulation and Implementation Results 69

4.3.1 Simulation Results 69

4.3.2 Implementation Results 72

4.4 Conclusion 77

CONTENTS vi

5 Mid-frequency Runout Compensation in HDDs via a Time-Varying

5.1 Introduction 84

5.2 Design of “Add-on” Linear Time-Varying Group Filter for Mid-f RRO Compensation 86

5.2.1 Group Filter Structure: Parallel Realization 87

5.3 Selection of Design Parameters 93

5.4 Application to a HDD servo system 97

5.4.1 Main Servo Compensator Design 97

5.4.2 Design of the LTV Group Filter for Mid-f RRO Compensation 98 5.4.3 Stability Analysis 99

5.4.4 Simulation and Implementation Results 100

5.5 Conclusion 107

6 An H∞ Disturbance Observer Design for High Precision Track Fol-lowing 116 6.1 Introduction 116

6.2 Proposed DOB Design 118

6.2.1 DOB with nominal feedback controller 118

6.2.2 H Q-filter design 121

6.3 Designs of nominal feedback controller C(z−1) and DOB 122

6.3.1 Simulation and implementation results 126

6.4 Conclusion 131

7.1 Findings and Conclusions 132

7.2 Suggestions for Future Research 134

of conventional HDDs in the global data storage systems market.

Nonetheless, as the multimedia world progressively enters into the high-definitionera, the demand of storage space increases dramatically for desktop and multime-dia applications With the latest developments in magnetic recording technologies,HDD manufacturers are producing HDDs of such a high storage density, which SSDsmanufacturers are incapable of achieving presently and in the near future This par-ticular advantage that conventional HDDs has over SSDs ensures that conventionalHDDs will possess a substantial market share for data storage systems in the fore-seeable future However, to gain a wider application in future data storage systems,future HDDs will have to offer greater storage capacities as well as faster data accesstime

The objective of this dissertation is to provide new, effective and practical trol algorithms to improve the performance of HDD and servo track writing (STW)servo system for the development of fast and high storage density HDDs In order

con-to increase the track density of our STW platform, an enhanced sensitivity tion shaping technique, which is based on the generalized Kalman-Yakubovic-Popov(KYP) lemma, has been developed As for improving the present HDD servo control

func-viii

technology, a unified control scheme (UCS) for high performance track-seeking andtrack-following in hard disk drives (HDDs) However, despite its effectiveness andsimplicity, the proposed UCS may not be appealing to the HDD industry, which

is generally conservative by nature, as it requires a HDD servo control design andimplementation overhaul Hence, two more new control algorithms which can bedesigned and implemented using an “add-o” fashion that improve different aspects

of the servo performance of existing HDD servo systems, while preserving the servoperformances as well as closed-loop stability margins achieved by the existing servodesign have been proposed too

List of Tables

2.1 Parameters of the STW platform 25

4.1 Comparison of operation numbers of the two control schemes 68

5.1 Design specifications 98

5.2 Comparison of stability margins at steady state 105

x

1.1 Components of a 3.5” typical HDD 41.2 Embedded servo scheme 71.3 An illustration of track seeking, track settling and track following of

a HDD servo system 81.4 Mechanical structure of a typical HDD and the block diagram of itsVCM-actuated servo loop system 91.5 Media-level servo track writing 15

2.1 Implementation setup of HDD servo system 202.2 Frequency responses of the VCM actuator (LDV range 2 µm/V) 222.3 Photo of a dissected 3.5” HDD 222.4 Overview of the STW platform with dual PZT microactuators 242.5 Zoom-in head-disk-spindle assembly of the STW platform 242.6 Servo mechanism of the MSTW setup (MicroE loop uses optical sens-ing signal as the feedback signal to a PID controller We focus on thecontrol design for PZT microactuator with readback PES as the feed-back signal 25

xi

LIST OF FIGURES xii

2.7 Frequency responses of the PZT microactuator 27

2.8 Measured RRO power spectrum of the STW platform prior to servo control 28

2.9 Measured NRRO power spectrum of the STW platform prior to servo control 29

3.1 A simplified block diagram of HDD servo loop with disturbance and noise injected 33

3.2 Case 1 : |S| and |T | 38

3.3 Case 1 : |TQ| 39

3.4 Case 2 : |S| and |T | 40

3.5 Case 2 : |TQ| 41

3.6 Case 3 : |S| and |T | 43

3.7 Case 3 : |TQ| 43

3.8 Multiplicative uncertainty ∆ of the microactuator 46

3.9 |S| and |T | Proposed scheme further reduces |S| at frequency < 200 Hz and maintains an equivalent low gain at around 650 Hz and 3.8 kHz while |T | at frequency > 1.5 kHz (esp after 15 kHz) improves instead of getting worse 47

3.10 Measured NRRO power spectrum with servo control Sudden appear-ance of peaks at higher frequency is due to the humps in the |S| from 1.2 to 7 kHz with the respectively servo controllers 48

3.11 |T WU| 48

4.1 A block diagram of a typical HDD servo system without any controller 53

4.2 Block diagram of the proposed NLTV control scheme 56

4.3 Internal stability analysis diagram of ˜xp 60

4.4 Nyquist plot of G1(s) 67

4.5 Simulation results: Normalized responses under a single conventionalCNF control law whose design parameters are optimized for the nom-inal 2 µm 70

4.6 Simulation results: Normalized responses under a single UCS controllaw whose design parameters are optimized for the nominal 2 µm 71

4.7 Simulation results: Nominal seeking performance robustness of theproposed scheme against plant uncertainty 71

4.8 CNF: 2 µm seek, Ch1: LDV-measurement (2 µm/V), r, Ch3: driver input, u, y(t), Ch4: Seek Command Input (2 µm/V) 72

VCM-4.9 UCS: 2 µm seek, Ch1: LDV-measurement (2 µm/V), r, Ch3: driver input, u, y(t), Ch4: Seek Command Input (2 µm/V) 73

4.10 UCS: 0.2 µm seek, Ch1: LDV-measurement (2 µm/V), r, Ch3: driver input, u, y(t), Ch4: Seek Command Input (2 µm/V) 74

4.11 UCS: 20 µm seek, Ch1: LDV-measurement (2 µm/V), r, Ch3: driver input, u, y(t), Ch4: Seek Command Input (2 µm/V) 75

VCM-4.12 Experimental results: Conventional CNF performance robustness against

±10% variations in the nominal seek length 76

LIST OF FIGURES xiv

4.13 Experimental results: Conventional CNF performance robustness against

±20% variations in the nominal seek length 77

4.14 UCS performance robustness against initial velocity of 3 mm/s Ch1:LDV-measurement (2 µm/V), Ch4: Trigger Indicates the start of anominal 2 µm seek 78

4.15 UCS performance robustness against initial velocity of 6 mm/s Ch1:LDV-measurement (2 µm/V), Ch4: Trigger Indicates the start of anominal 2 µm seek 79

4.16 UCS performance robustness against initial velocity of 12 mm/s Ch1:LDV-measurement (2 µm/V), Ch4: Trigger Indicates the start of anominal 2 µm seek 79

4.17 UCS performance robustness against initial acceleration of 4.5 ×103

mm/s2 Ch1: LDV-measurement (2 µm/V), Ch4: Trigger Indicatesthe start of a nominal 2 µm seek 80

4.18 UCS: Measured nominal 2 µm step response under the effect of turbance and noise Ch1: LDV-measurement (2 µm/V), Ch4: SeekCommand Input (2 µm/V) 80

4.19 CNF: Measured nominal 2 µm step response under the effect of turbance and noise Ch1: LDV-measurement (2 µm/V), Ch4: SeekCommand Input (2 µm/V) 81

dis-4.20 Measured frequency response of open-loop function (LDV range 2µm/V) 81

4.21 Measured frequency gain response of sensitivity function 82

4.22 Power spectrum of measured PES NRRO with servo control 82

5.1 Open-loop block diagram of a typical HDD servo loop sated with a main servo compensator 86

precompen-5.2 Block diagram of the proposed LTV group filter, Ft, added onto themain feedback loop via parallel realization 87

5.3 Maps of eigenvalues of A0

cl+A∞

∆ ‘×’: Nominal plant; ‘◦’: Resonancesvariations: -10% frequency and damping shift; ‘+’: Resonances vari-ations: +10% frequency and damping shift 100

5.4 Simulated disturbance responses at 700 Hz Approx transient tling time, Ft: 3 ms; FLT I1(z): 2.5 ms; FLT I2(z): 6.5 ms 101

set-5.5 Simulated disturbance responses at 2 kHz Approx transient settlingtime, Ft: 1.8 ms; FLT I1(z): 1.5 ms; FLT I2(z): 6.5 ms 102

5.6 Simulated combined disturbance responses at, respectively, 2 kHz and

700 Hz Approx transient settling time, Ft: 3 ms; FLT I1(z): 2.5 ms;

FLT I 2(z): 6.5 ms 102

5.7 Simulated sensitivity function magnitudes with respective RRO pensation scheme at the steady state Note: The notch widthswith Ft are equally wide as those with FLT I1(z) and theirsensitivity function humps are equally sized at t = 0 103

com-5.8 Simulated open-loop frequency response with respective RRO pensation scheme at the steady state 104

com-5.9 Proposed LTV method, Ft: Measured disturbance responses at 700

Hz Ch1: LDV-measurement (2 µm/V), y, Ch2: VCM-driver input,

u Approx transient settling time: 3 ms 108

LIST OF FIGURES xvi

5.10 Proposed LTV method, Ft: Measured disturbance responses at 2 kHz.Ch1: LDV-measurement (2 µm/V), y, Ch2: VCM-driver input, u.Approx transient settling time: 1.5 ms 108

5.11 Proposed LTV method, Ft: Measured disturbance responses at 700and 2 kHz Ch4: LDV-measurement (2 µm/V), y, Ch3: VCM-driverinput, u Approx transient settling time: 3 ms 109

5.12 Conventional LTI method, FLT I2(z): Measured disturbance responses

at 700 Hz Ch1: LDV-measurement (2 µm/V), y, Ch2: VCM-driverinput, u Approx transient settling time: 7 ms 110

5.13 Conventional LTI method, FLT I2(z): Measured disturbance responses

at 2 kHz Ch1: LDV-measurement (2 µm/V), y, Ch2: VCM-driverinput, u Approx transient settling time: 6 ms 110

5.14 Measured sensitivity functions with respective compensation scheme

at the steady state 111

5.15 Measured open-loop frequency responses with respective tion scheme at the steady state (LDV range 2 µm/V) 111

compensa-5.16 Simulated sensitivity functions with respective compensation scheme

at the steady state 112

5.17 Simulated disturbance responses of 0.4 µm at 700 Hz Approx sient settling time, Ft: 3 ms; FLT I1(z): 2.5 ms; FLT I2(z): 6.5 ms 112

tran-5.18 Proposed LTV method, Ft: Measured disturbance responses of 0.4

µm at 700 Hz Ch4: LDV-measurement (2 µm/V), y, Ch3: driver input, u Approx transient settling time: 3 ms 113

VCM-5.19 Simulated disturbance responses of 0.1 µm at 3 kHz Approx tran-sient settling time, Ft: 0.9 ms; FLT I1(z): 0.7 ms; FLT I2(z): 10 ms Note: Ft= F3

t + F4

t 113

5.20 Simulated sensitivity functions with respective RRO compensation scheme at the steady state Note: Ft= F3 t + F4 t 114

5.21 Proposed LTV method, Ft: Measured disturbance responses of 0.1 µm at 3 kHz Ch4: LDV-measurement (2 µm/V), y, Ch3: VCM-driver input, u Approx transient settling time: 1 ms Note: Ft= F3 t + F4 t 114 5.22 Simulated disturbance responses of 0.4 µm at 1 kHz Approx tran-sient settling time, Ft: 2.3 ms; FLT I1(z): 2 ms; FLT I2(z): 4 ms Note: Ft = F3 t + F4 t 115

6.1 Block diagram of DOB with the nominal feedback controller 119

6.2 Approximately simplified block diagram of Fig 1 120

6.3 Frequency responses of respective Q-filters 125

6.4 Illustration of attenuation of nc with DOB using Nyquist plots of (Qconv− 1) and (Qprop− 1) 126

6.5 Nyquist plot of Tolin (6.20) for the nominal and perturbed plant with Qprop.Gain margin: 8.5 dB (Nominal), 6.5 dB (−10 % perturbation), 7 dB (+10 % perturbation); Phase margin: 50 deg (Nominal), 45 deg (−10 % perturbation), 47 deg (+10 % perturbation) 127

6.6 Simulated sensitivity gain for the nominal and perturbed plant with Qprop. 128

6.7 Open-loop frequency responses with the nominal feedback controller C without DOB 128

LIST OF FIGURES xviii

6.8 Frequency responses of Tolwith DOB designed using the conventionalmethod, Qconv. 129

6.9 Frequency responses of Tol with DOB designed using the proposedmethod, Qprop. 129

6.10 Measured gain of the sensitivity transfer function 130

6.11 Power spectrum of measured PES NRRO with respective methods 131

7.1 Photo of a dual-stage actuator with a PZT-actuated suspension 136

Throughout this thesis, the following abbreviations and notations, which are fairlystandard, are adopted.

ARE Algebraic Riccati equation

BPI Bits Per Inch

CNF Composite Nonlinear Feedback

DOB Disturbance Observer

FFT Fast Fourier Transform

FIR Finite Impulse Response

FPC Flexible Printed Circuit

GES Global Exponential Stability

G-KYP Generalized Kalman-Yakubovich-Popov

GUS Global Uniform Stability

GUAS Global Uniform Asymptotic Stability

HDD Hard Disk Drive

HAMR Heat Assisted Magnetic Recording

IIR Infinite Impulse Response

IVC Initial Value Compensation

KYP Kalman-Yakubovich-Popov

LMI Linear Matrix Inequality

LDV Laser Doppler Vibrometer

xix

NOMENCLATURE xx

LQG Linear Quadratic Gaussian

LTI Linear Time-invariant

LTR Loop Transfer Recovery

LTV Linear Time-varying

MISO Multi-Input, Single-Output

MSC Mode Switching Control

MSTW Media-level Servo Track Writing

NLTV Nonlinear Time-varying

NRRO Nonrepeatable Runout

NLSO Nonlinear Least Squares Optimization

NLTV Nonlinear Time-varying

PES Position Error Signal

PID Proportional-Integral-Derivative

PMR Perpendicular Magnetic Recording

PTOS Proximate Time Optimal Servomechanism

SSD Solid State Disk/Drive

STW Servo Track Writing

TMR Track Misregistration

TPI Track Per Inch

UCS Unified Control Scheme

VCM Voice Coil Motor

ZOH Zero Order Hold

1.1 Background and Motivation

In November 2007, Samsung, the world’s top company in global electronic industrylaunched a new generation of solid state disks (SSD) that used native SATA-IIinterface [1] Those new SSDs were coming in 1.8” and 2.5” form factor Theyfeatured 960 Mbps sequential read and 800 Mbps sequential write speed Theirpower consumptions were rated at a mere 0.7W Those were the very first SSDs,

in the entire storage industry, that not just rivaled, but surpassed conventionalhard disk drive (HDD) in read/write (R/W) speed And in the case of 1.8” formfactor, the SSDs matched storage capacity of its HDD counterpart as well It hadbeen reported that such high performance alternative storage devices threatenedthe future presence of conventional HDD as the leading product group in the globaldata storage systems market

Major HDD manufacturers, namely Seagate Technology, Western Digital nology as well as Hitachi Global Storage Technologies, are finding it hard to faceoff the SSD in term of performances in power consumption, random memory accesstime, robustness against shock and production of heat and acoustic noise As a

Tech-1

CHAPTER 1 2

result, the deployment of SSDs in mobile applications, in which power tion and resistance against shock disturbances are serious matters of concern, hasbeen very successful over the last few years Global computer makers such as AppleComputers, Samsung and DELL Computers are already selling their latest high-endnotebook personal computers (PC) that use SSDs According to one of the latestreports produced by IDC, the premier global market intelligence firm, it is pre-dicted that SSDs will go mainstream as advances in solid state memory technologyand dipping price points drive worldwide SSD adoption for mobile applications [2].Going forward, the PC market presents the biggest demand for storage devices andthe PC market is transitioning from one being dominated by desktop PC shipments

consump-to one being dominated by notebook PC shipments This transition increases theimportance of low power consumption and shock resistance requirements, which aredynamics that align very well with the benefits of SSDs

While HDD manufacturers are still unable to produce HDDs that match SSDs

in general performance, they are still capable of ever-increasing the storage density

of HDD and have already packed as much as 1 Terabytes (TB) of data into a 3.5”HDD [3] by using the latest perpendicular magnetic recording (PMR) technology [4]that significantly increases bit density, which is measured by bits per inch (BPI) Asthe PMR technology matures, higher BPI, consequently higher storage density, will

be achieved The development of newer recording technologies such as heat assistedmagnetic recording (HAMR) [5] as well as patterned media magnetic recording [6]are in the progress Once completed, they will be ready to replace PMR and pushingfor even higher BPI as well as and hence, higher storage density By far, the rate ofincrease of storage density of HDD is still ahead its SSD counterpart [7]

As the multimedia world progressively enters into the high-definition era [8], theincreasing demand for even more storage space for desktop and home multimediaapplications has already become the norm [9] This ensures that HDD will possess asubstantial market share for desktop and home multimedia applications Moreover,for these applications, other than having a faster random access memory R/W speed

and lower acoustic noise production, the rest of the advantages that SSDs have overthe conventional HDDs, such as lower power consumption and shock resistance, arerelatively less relevant.

However, though HDDs already have gained a strong foothold in desktop andhome multimedia applications, to gain a wider application in future data storagesystems, future HDDs will have to offer larger storage capacities as well as fasterdata access time

While new recording media and technologies are being consistently developed

to achieve ever higher BPI, the BPI is anticipated to reach the upper limit soon,which is the result of thermal instability of the recording media Thus, to avoid thethermal degradation in magnetic recording, higher storage density can be achieved

by pushing for higher track density, measured by track per inch (TPI) As for a nificant reduction in data access time, track seeking times achieved using boundedcontrol effort have to be shorten In summary, to assist HDDs to maintain a signifi-cant market share for desktop and home multimedia storage applications in future,new control strategies will need to be formulated to achieve higher track followingaccuracy as well as faster and smoother track seeking motion This dissertation isconcerned with the development of new, effective and practical control algorithmsfor fast and high storage density future HDDs

sig-In the rest of this chapter, to assist better understanding of the HDD servotechnology, brief information about the history and present of HDD, servo trackwriting (STW) platform as well as their respective servo systems, will be studied

1.2 Hard Disk Drives

A hard disk drive (HDD) is a non-volatile storage device which records digital dataonto rotating disks whose surfaces are coated with a magnetic material Presently,

CHAPTER 1 4

HDDs are the most prevailing devices used for large and permanent data storage.During the past half a century, HDD technology has made dramatic progress interms of physical size, storage capacity, R/W speed and cost While the first HDDs

of an 8” form factor had a capacity of 10 Megabytes (MB), costing over $100 per

MB, nowadays, a 3.5” HDD of 500 Gigabytes (GB) costs less than half a dollarper GB In the mean time, the performance, reliability as well as the data accesstime and transfer rate have been improved drastically HDDs are also available in avariety of sizes to widen its applications in data storage systems The smallest HDDyet [10], which is of a 0.85” form factor and comes with 8 GB of storage capacity,had been developed by Toshiba in 2004 was to be used in ultra-portable consumerelectronics devices

Figure 1.1: Components of a 3.5” typical HDD

The components of a typical 3.5” HDD are shown in Fig 1.1 [11,12] As shown

in Fig 1.1, HDD contains a stack of round, flat disks called disks or disk platters,which are coated with a layer of magnetic material on their surfaces Information isrecorded on the magnetic coating of the disk platters This stack of disks is stackedalong a spindle and rotated by an electric motor inside the spindle, which is calledthe spindle motor Electromagnetic R/W heads are used to read or write digital

information from or onto the disks These heads are mounted on the suspensions andthe suspensions are attached to Voice coil motor (VCM)-actuated actuator arms.The flexible structure of the suspensions helps to maintain a relatively constantflying height on a thin layer of air bearing over the rotating disks While the flyingheight should be kept as low as possible to ensure the maximum signal-to-noiseratio (SNR) R/W operation, during R/W operation, the heads should never come

in contact with the surface of the rotating disks Else, a head crush is said to haveoccurred and the heads and the disk surfaces would have been seriously damaged

On the disk surfaces, data are recorded in tightly-packed concentric rings calledtracks An increase of HDD storage capacity can be achieved by the ability to recordmore bits of information onto a single disk, i.e to increase the areal density of thedisks Areal density is measured in bits per square inch (BPSI), which is given by

where BPI stands for the bits per inch along any single track and TPI stands fortracks per inch Thus, higher BPSI can be achieved by either increasing BPI orTPI, or even both

Areal density is an important factor of HDD performance Increasing TPI is arelatively easier approach to increase BPSI, however, it is still a challenging task.Historically, it involves cross-disciplinary technological advances in magnetic media,actuators and disk design, servo algorithm, and microprocessor or DSP This disser-tation emphasizes the development of new servo algorithms to improve HDD servosystem

is embedded in the servo sectors along the data tracks, which are at spaced equally inradial direction, with data sector, in which user data information are recorded onto,

in between By demodulating and decoding the information inside a servo sector,information such as track number, or track ID as well as the distance between of headposition and track center, i.e position error signal (PES), which is encoded in thePES burst, can be obtained These servo sectors are pre-written onto the disks beforethey can be used The process of writing servo sectors is known as servo writing

It is performed using a STW platform And because the HDD servo system usesthe servo information on the servo sector as position sensor, consequently, the HDDservo performance is heavily dependant on the quality of the STW process [13, 12].Thus, to achieve an excellent HDD servo performance, the servo system of STWplatform must attain a much higher positioning accuracy than HDD servo system

In an embedded servo system, the PES can only be obtained in discrete-timeformat Thus, its servo system is discrete in nature The product of the spindlespeed and the number of sectors determines its sampling rate, which is given by

Figure 1.2: Embedded servo scheme.

1.3.2 Servo Control

The HDD servo loop is illustrated in Fig 1.4 In general, the HDD servo systemrefers to the VCM actuator servo system [14], or the HDD R/W head positioningmechanism As seen in Fig 1.4, there is a servo controller It is designed in such

a way that the HDD servo system is capable of its three main modes of operation,namely, the track seeking mode, the track settling mode and the track followingmode, ideally under the effect of noises, disturbances and plant variations Fig 1.3shows a graphical illustration of various modes of operation of a HDD servo system

The respectively details, such as the purpose as well as the related servo controlalgorithms, of each mode of operation are given as follows

A Track Seeking Mode

The control objective during the track seeking mode aims to move the R/W fromany track to the desired track, with bounded control effort, in a fast and smooth

be more effective than PTOS By cancelling most unwanted dynamics of the loop system with a stable feedforward controller, 2-DOF is capable of achieving veryfast and smooth track seeking performance without affecting stability as well asnoise and disturbance attenuation capability of the closed-loop system While theidea behind 2-DOF is simple and straightforward, its implementation is difficult,especially when some of all the unwanted dynamics are non-minimum phase innature Various control designs such as the zero-phase error track control (ZPETC),which is a single-rate control method, and perfect tracking control (PTC) [17], which

closed-Figure 1.4: Mechanical structure of a typical HDD and the block diagram of itsVCM-actuated servo loop system.

CHAPTER 1 10

is multi-rate control method, have been developed to compensate non-minimumphase unwanted dynamics satisfactorily

B Track Settling Mode

Once the R/W heads have reached the target track, the HDD servo system willchange its mode of operation to track following A fast and smooth operation modetransition will guarantee minimum time delay for fast data access However, duringmode switching, the states variables of VCM actuator, that is our plant, are usually

at some non-zero positions Given the fact that most VCM have very bad andunwanted dynamics, the non-zero states of the plant will ensure the mode switchingprocess to be slow and jerky

In 1996, Yamaguchi et al developed a novel method called initial value pensation (IVC) [18–20] to compensate the undesirable effects of the non-zero plantstate variables during mode switching Initially, the design for IVC control methodsaims to cancel the undesirable dynamics of responses owing to non-zero plant statevariables upon switching In its later development, optimal control theories [21]have been applied to enhance the performance of IVC

com-Recently, a new nonlinear control technique, the composite nonlinear feedback(CNF) control [22], has been developed and it eliminates the need to compensatethe effect of non-zero plant variables during mode switching A single control rule

is designed as a unified control scheme for track seeking and track following Byhaving a linear feedback gain to ensure a fast response and a nonlinear gain thatvaries the damping ratio of the closed-loop system appropriately according to trackerror, the CNF shows it is capable of achieving a fast and smooth track seekingperformance that surpasses that of PTOS

C Track-Following Mode

The goal of the HDD servo system during the track following mode is to keepthe R/W head as close as possible to the track centre to ensure maximum SNRduring magnetic reading and writing Specifically, this implies, under the effect ofvarious noises and disturbances as well as plant variations within the HDD servosystem [11, 12], the track following servo controller needs to minimize the trackmisregistration (TMR), which is a measurement of track following performance and

is given by

where σpes is the standard deviation of PES And for an acceptable SNR level ofmagnetic reading and writing, TMR must be controlled within 10% of a track pitch[11, 12] This implies, given any TMR, the maximum achievable track density can

disk-Till today, various control strategies have been introduced to reduce the effect

of noises and disturbances, and thus TMR While the feedforward control methods

CHAPTER 1 12

are effective against external disturbances as well as for the identified RRO, close

to 50% of the TMR [13] can only be reduced via a well designed feedback servocontroller The servo control technologies that have been developed over all theseyears can be roughly categorized as follows

1 Classical Loop Shaping

Loop shaping designing methods based on classical frequency domain techniquessuch as notch filtering [32–34], resonance/peak filtering [35–38], lead-lag phase com-pensators [14, 39, 40], provide more intuitions and a greater ability to tune designs

to achieve desired frequency domain properties for the closed-loop system than tomated tools based on state-space formulation

au-2 Advanced Optimal Control

Advanced model-based controller designing methods based on state-space tion such as LQG/LTR [41, 42, 11], H2 [11, 32], H∞ [11, 14, 43, 44], mixed H2/H∞

formula-[45–48] as well as disturbance decoupling or almost disturbance decoupling are pable of minimizing any pre-defined frequency domain cost function, some withstability and performance robustness consideration against plant variations Thus,only with these advanced methods, it is theoretically possible to achieve the highestpossible theoretical TPI Such a task is virtually impossible when using the classicalloop shaping method

ca-3 Adaptive and Self-Tuning Controller

Owing to the fact that plant parameters as well as stochastically-static disturbanceand measurement noise models may vary under environment changes, it is theoret-ically impossible to realize the maximum disturbance and noise attenuation perfor-mance with a fixed controller at all times Thus, various adaptive and self-tuning

control algorithms [49–53] have been developed to achieve robust performance aswell as stability.

4 Multi-sensing Servo Control

While conventional HDD servo system achieves servo information with only thePES feedback signal, augmenting the servo loop with additional sensors provide theopportunity to having extra degrees of freedom in controller design so as to achievebetter servo performance Several new control algorithms [54, 49, 55–58] that makeuse of the additional servo information providing by the additional sensors have beendeveloped recently

5 Multi-rate Servo Control

Similar to multi-sensing servo control, multi-rate servo control provides extra grees of freedom in controller design Ignoring the intersample behaviors, [59] showsthat it is mathematically possible to always obtain a minimum-phase system afterdiscretizing a continuous plant at any arbitrary sampling period by making use ofmultirate sampling Besides that, [60] displays capability of achieving disturbancesuppression beyond the nyquist frequency, which is limited by the spindle speed aswell as the number of servo sectors on the disks, as discussed earlier

de-6 Nonlinear Control

Currently, prevailing HDD servo controllers are linear Previous studies have shownthat there are various performance limitations and tradeoffs associated with linearcontrol systems Bode’s gain-phase relationship [61], Bode’s integral theorem [62]and time-domain constraint [63] are some of the famous ones whose effects are thefarthest-reaching as well In recent years, novel nonlinear control methodologies [11,

CHAPTER 1 14

64,65] have been introduced and developed They have shown that they are capable

of bringing substantial improvement to current standard of HDD servo performance

by means of breaking the performance limitations of linear control systems

1.4 Servo Track Writing Technology

As mentioned earlier, the servo performance of a HDD depends heavily on the quality

of its servo tracks, which provide the head positioning information Till date, threedifferent servo patterns have been developed and used in HDD They are, namely,wedged servo, dedicated servo and embedded servo Presently, the commonest diskservo pattern is the embedded servo pattern, which reserves most disk areas to

be allocated for data recording as well as being immune to mechanical offsets inactuator assembly [12, 66]

Time taken to write a complete servo pattern on a single disk is the throughputtime Since servo patterns are written track-by-track, as TPI increases, throughputtime increases as well In order to reduce throughput time so as to reduce produc-tion cost, HDD industry have developed a media-level servo track writing (MSTW)technology, which servo writes multiple disks instead of a single disk within the sameperiod of time Fig 1.5 illustrates MSTW process The MSTW platform is made

up of an air-bearing spindle motor, which holds all the magnetic disks Like inside aHDD, there is a VCM actuator with multiple magnetic heads The servo controller

of the actuator uses external devices to provide position information of the headswith respect to the stack of disks

1.5 Contributions of the Thesis

As mentioned earlier in the introduction, new and effective servo control algorithmsneed to be developed to help HDD to maintain its leadership in market share of

Figure 1.5: Media-level servo track writing.

data storage for desktop and multimedia applications This research will do justthat It will provide new servo solutions to help to improve the head positioningaccuracy of STW platform, which consequently, improves the servo performance

of HDD, as mentioned in the previous section A novel nonlinear time-varying(NLTV) control algorithm is developed, which will be used to design a unified controlscheme that achieves great track seeking, track settling as well as track followingperformance While the proposed unified scheme requires a major HDD servo designand implementation overhaul, the thesis also has come up with two new “add-on”solutions These new “add-on” servo solutions are capable of enhancing any existingHDD servo system while preserving the servo stability margins of the main feedbackloop The contributions made by this thesis are summarized below:

1 A new servo control design for the development of a high TPI MSTW form has been proposed Modifications have been proposed to enhance theoriginal G-KYP lemma based sensitivity function S(z) shaping technique thatQ-parameterizes the controller and solves for the desired finite impluse re-sponse (FIR) filter Q(z) using Linear Matrix Inequalities (LMI) optimization

plat-By representing Q(z) with an infinite impulse response (IIR) filter and ing an extra LMI that is derived based on the Bounded Real Lemma into theoriginal LMI optimization algorithm, problems such as constraints in selection

includ-of design specification, degradation in noise rejection and insufficient stabilityrobustness against plant uncertainty are effectively alleviated In other words,the proposed scheme aims to achieve a better compromise between distur-

fol-to achieve fast and smooth seeking In order fol-to be advantageous over theconventional CNF control, the UCS consists of both linear and nonlineartime-dependent components, which are independent of absolute seeking error.During track seeking mode, the proposed scheme has a significantly betterperformance robustness against variations in seek length as compared to theconventional CNF control For track following, disturbance and noise modelsare involved in the problem formulation and sub-optimal H2 method is used

to design the controller so as to achieve high head-positioning accuracy

3 Conventional “add-on” feedback filters that are designed to compensate formid-frequency (mid-f ) RRO, which are of known center frequencies, but un-known magnitudes and phases, in a HDD servo system either have a long filtertransient or constitute a large sensitivity hump as well as poor stability mar-gins In this dissertation, a novel linear time-varying (LTV) group filteringscheme for the compensation of several mid-f RRO harmonics is presented.While having a short filter transient that ensures fast disturbance attenuation,the proposed filter does not constitute to any substantial unnecessary sensi-tivity gain amplification at the steady state Simulation and implementationresults show the effectiveness of the proposed group filtering scheme used tocompensate two mid-f RRO harmonics simultaneously

4 An alternative approach of designing an add-on disturbance observer to hance general low frequency disturbances attenuation Contrary to the conven-tional method, instead of designing Q-filter to be a low-pass filter with unity

en-DC gain, the optimal H∞ method has been applied The proposed method iscapable of removing more low frequency disturbances while not compromising

the excellent disturbances attenuation performance at higher frequency regionachieved by the nominal feedback loop.

1.6 Organization of the Thesis

This chapter has given a general overview of the HDD’s future, the HDD servosystem as well as the STW technology The research motivation and objectiveshave been discussed as well

Chapter 2 discusses the details of the modeling of VCM actuator of HDD aswell as the Lead-Zirconium-Titanate (PZT) actuator A detailed description of ourexperimental platform will be provided in this chapter as well

Chapter 3 presents the enhanced G-KYP Lemma based sensitivity functionshaping technique that has been used to increase head positioning accuracy of ourMSTW platform

In Chapter 4, a novel NLTV unified HDD control scheme for seeking and trackfollowing has been developed Its designing method and stability issues are detailed

in this chapter

Chapter 5 proposes a novel add-on LTV group filter, which can be used toenhance mid-f RRO attenuation performance of any existing HDD servo system,while preserving the servo stability and performance of the original HDD system

Chapter 6 presents an alternative method of designing the Q-filter of an add-ondisturbance observer that helps to improve the disturbance attenuation capability

of any present HDD servo system for general type, low-frequency disturbance

Finally, Chapter 7 concludes the thesis Suggestions for future works are cussed here