2D Coding andIterative Detection Schemes
J. A. O’Sullivan, N. Singla, Y. Wu, and R. S. Indeck
Washington UniversityMagnetics and Information Science Center
Nanoimprinting and Switching of Patterned Media
A Washington University –Rowan University
Collaboration
ss
Enabling Technology: Disk Drives
Magnetic disk storage areal density vs. year of IBM product introduction
(From D. A. Thompson)
~10,000,000x increase in 45 years!
Conflicting Demands on Media
• Decrease volume to increase storage density
• Maintain volume for thermal stability
Medium noise is less in a medium with smaller grains
intended location
actual location
Patterned Media: 1 bit/’grain’
• Increase magnetic region for stability and low noise– increase intergranular exchange
• Create single domain regions– switching (no wall motion)– predefine placement
Motivation
Need medium which can sustain smaller, stable magnetic regions1. Continuous film with discrete ‘grains’; echange ds-coupled;
make grains smaller; create bits from 1000 grains2. Discrete film; glue (via exchange) structure within region; 1 bit
per region2a. Smooth film, ion beam medification, create discrete regions, mag.
w/in region exchange coupled
Want a manufacturable process!
Viable Patterned Medium Fabrication Technique
• Grow magnetic material• Spin on sacrificial polymer• Nanoimprint polymer at room-temperature• Ion beam modify material to produce
islands of magnetic material• Remove polymer
Challenges with thePatterning of Media
• Fine features• Large area coverage• Complicated, non-rectangular structures• High substrate throughput
. . . conventional processing inadequate
Imprint Lithography
Nanoimprint Lithographyfor patterning recording media
bilayerresist
metal
(d)
(e)
(f)residual polymer
imprinter
substrate(a)
(b)
(c)
Nanoimprintingfor patterning recording media
Room-TemperatureNanoimprint Lithography
• A hyperbranched perfluorinated polymer (HBFP) is the RT imprint resist
• The low glass transition temperature (Tg=54°C) allows RT resist displacement
• Have demonstrated fine features (<20 nm) of imprinting using Alumina imprinter
Ion Beam Modification
Terris: IBM
Nanoimprinting/Ion Modification
• Small-scale features (10 nm)• Large-scale extensibility• Smooth topography• Suitable for irregular patterns
– non-rectangular, sector and servo
. . .viable for patterning recording media, memory elements and sensors
Nanoimprinting
6 million-million
bits/square inch!
Medium Microstructure and Noise Each microscopic region has uniaxial
anisotropyRandomness arises from the medium microstructure Dominant noise source in media
Enhanced MRM Image
H
M
Magnetization Change on theMicroscopic Level
MFM Images at Various Reversal Stages
Happ = -6400 Oe Happ = 1220 Oe Happ = 1350 Oe Happ = 1550 Oe Happ = 6400 Oe
MFM Images of Positive and Negative Saturation
Medium Noise Experiment
downtrack position (µm)
0 10 20 30 40 50
MR
read
back
sig
nal
-1
0
1
zoom in
Determinism of Medium Noise
downtrack position (µm)
0 2 4 6 8 10
MR
read
back
sig
nal
positive saturationnegative saturation
Simulated Medium
Magnetic Transitions
Superparamagnetism
• Two types of materials:– Those that have a net magnetic moment at each lattice
site and those that don’t• Ferromagnetic material’s moments stick together
(exchange)• Temperature can jumble up the moments
(paramagnetism)Small volumes can spontaneously, coherentlychange (superparamagnetism)
Time Dependence of Magnetization
dN1 2dt
N1(t)e
E1kT→ ∝
−∆
(b)
Time Dependence of Magnetization
M(t) NM e NM where = C er
t
r-1
EkT= −
−2 τ τ
∆
Single Activation Energy Activation Energy Distribution
( )M(t) 2M NkTp lnttro
= −⎛⎝⎜
⎞⎠⎟
M(t) Slnt + const=
∆E = KV 1-HH
a
k
⎛⎝⎜
⎞⎠⎟
2
f(H )k
Hk
f(H )k
Hk
Time Dependence of Magnetization
Spin-standMFM, VSM
Dynamic
10-13 10-9-10-8 10-4 1
time (seconds)
Mag
netiz
atio
n
time (seconds)
10-1 100 101
Nor
mal
ized
Am
plitu
de
0.94
0.95
0.96
0.97
0.98
0.99
1.00
1.01
1500 frmm3000 frmm4550 frmm
300 K
Short-term Experiment
Stability of Magnetic Recordings
300 K
330 K
341 K
355 K
time (seconds)
10-1 100 101
Nor
mal
ized
Am
plitu
de
0.88
0.90
0.92
0.94
0.96
0.98
1.00
1.02
300 K
341 K330 K
355 K
3000 frmm
Short-term Experiment
Stability of Magnetic Recordings
Temperature( K)
280 300 320 340 360 380 400 420
Coe
rciv
ity (O
er)
1000
1500
2000
2500
Effect of Temperature on Medium Properties
• Ms does not change• Reversible change in Hc : decreases by ~6 Oe per K
-4000 -2000 0 2000 4000-3
-2
-1
0
1
2
3x 10
-3
Applied Field (Oe)
Mag
netiz
atio
n (e
mu)
295 K343 K364 K384 K409 K
Presentation Outline
• Introduction• PRML channels• soft decoding channels
• Decoding for 2-D ISI channels• introduction of 2-D ISI• MMSE equalization and decoding• message passing on combined LDPC and ISI graph
• Decoding for separable 2-D ISI channels• separating 2-D ISI by equalization• decoder diagrams and performances
Transition Response
0 20 40 60 80 100 120 140 160 180 200-0.02
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
Am
plitu
de
Spinstand Captured Transition Response
Data Format
0 1 0 0 0 2 0 0 0 3 0 0 0 4 0 0 0 5 0 0 0 6 0 0 0 7 0 0 0 8 0 0 0 9 0 0 01 8 0
1 9 0
2 0 0
2 1 0
2 2 0
2 3 0
2 4 0
2 5 0
2 6 0
0 5 0 1 0 0 1 5 0 2 0 0 2 5 0 3 0 0 3 5 0 4 0 01 9 0
2 0 0
2 1 0
2 2 0
2 3 0
2 4 0
2 5 0
0 5 0 1 0 0 1 5 0 2 0 0 2 5 0 3 0 0 3 5 0 4 0 01 8 0
1 9 0
2 0 0
2 1 0
2 2 0
2 3 0
2 4 0
2 5 0
Preamble Synchronization Marker
PRML System
ReceiveFilter
(low pass)Adaptive
FilterViterbi
DetectorMagneticRecordingChannel
Timing &Tap Updating
Wide-BandNoise
nT+τe
ny nxnx ny
,....3,2,1 ,)1)(1()( =+−= nDDDP n
Serial Turbo Detector
DEMUX
Depuncture
APP
Outer
2π
xext
L
( )1
kpΛ
( )k
uΛ
−
1
2
−πAPP
Inner
⊕( )
kxΛ
−
kyMUX
Puncture⊕
I
OI
O
I
I
O
O
ku
Drive Data Experiment
5 6 7 8 9 10 1110-8
10-7
10-6
10-5
10-4
10-3
10-2
10-1
SNR [dB]
Bit
Erro
r Rat
e
Rate 16/17 ideal turboRate 16/17 drive turboIdeal EPR4 Drive EPR4
Advanced Media
Science enables:
6 million-million bits/square inch!
2-D Intersymbol Interference
x11 x12
x22x21
x13
x23
x14
x24
x34x33x32x31
x41 x42 x43 x44
-1 -1 -1 -1-1
-1
-1
-1 -1 -1 -1-1-1
-1-1-1-1-1
-1-1
r11 r12
r22r21
r13
r23
r14
r24
r34r33r32r31
r41 r42 r43 r44
r02 r03 r04 r05
r15
r01
r25
r52 r53 r54 r55
r45
r35
r00
r10
r20
r30
r040
r51r50
⎟⎟⎠
⎞⎜⎜⎝
⎛=
25.05.05.01
h ⊕
w(i,j)
jijijijijiji wxxxxr ,1,11,,1,, 25.05.05.0 ++++= −−−−
GUARD BAND
MMSE Equalization
• Combination of equalization and decoding for 2-D ISI
• Equalizer designed subject to peak power constraint
a xLDPCEncoder
a(i,j) x(i,j) r(i,j) MMSEEqualizer
LDPCDecoder
ChannelISI
w(i,j)
),( jia∧
),( jix∧
Channel
Iterative MMSE Equalization
• Soft information passed from LDPC decoder to equalizer
• Scheduling: MMSE-LDPC iterations
a xLDPCEncoder
a(i,j) x(i,j) r(i,j) MMSEEqualizer
LDPCDecoder
ChannelISI
w(i,j) Extrinsic Information
),( jia∧
),( jix∧
Channel
PerformancePerformance of MMSE Equalization and Decoding
-6
-5
-4
-3
-2
-1
0
1 1.5 2 2.5 3 3.5 4 4.5
SNR (dB)
Bit-
Erro
r R
ate
in lo
g10
ISI-free
ItrWiener 1-20
ItrWiener 1-10
ItrWiener 1-5
ItrWiener 1-1
Wiener
Full-graph Message Passing
• Message passing on the 3-level graph of the LDPC code and channel ISI
• Messages passed are probabilities • Algorithm computes a-posteriori probabilities
of the codeword bits given the observed data
LDPC Bipartite Graph
Check Nodes
Variable Nodes (x)
Full-graph
x(i+2,j)
x(i+1,j)
x(i,j)
x(i+2,j+1)
x(i+1,j+1)
x(i,j+1)
x(i+2,j+2)
x(i+1,j+2)
x(i,j+2)
r(i+1,j+1)
r(i,j+1)r(i,j)
r(i+1,j)
FromCheckNodes
Performance
Performance of Iterative Decoding Schemes
-6
-5
-4
-3
-2
-1
0
1 1.5 2 2.5 3 3.5 4 4.5
SNR (dB)
Bit-
Erro
r R
ate
in lo
g10
ISI-freeItrWiener_10 1-20Fullgraph_10Wiener
Full-graph analysis
• Length 4 cycles present which degrade performance of message passing algorithm
x(i+2,j) x(i+2,j+1)
x(i+1,j) x(i+1,j+1)
x(i,j) x(i,j+1)
x(i+2,j+2)
x(i+1,j+2)
x(i,j+2)
r(i+1,j+1)
r(i,j+1)r(i,j)
r(i+1,j)
FromCheckNodes
Ordered Subsets
• Taken from imaging applications – data sets grouped into subsets to speed up computations
• Eliminate loops in the channel ISI graph
Grouped ISI Graph
Performance
Performance of Iterative Decoding Schemes
-6
-5
-4
-3
-2
-1
0
1 1.5 2 2.5 3 3.5 4 4.5
SNR (dB)
Bit-
Erro
r R
ate
in lo
g10
ISI-freeOrdered SubsetsItrWiener 1-20FullgraphWiener
Consider Separating the ISI
Down-track separable from cross-track
Linear combination
Separating 2-D ISI
•Advantages of separating equalization•shortening ISI and reducing detector complexity•separating 2-D ISI and reducing detector complexity
w(i, j)
2-D ISIChannel
r(i, j)
w(i,j)
Vertical ISI Horizontal ISI
r(i, j)y(i, j)
Separating Equalization
Separable channel response
A Separable 2-D ISI
( )5.015.0
125.05.05.01
⎟⎟⎠
⎞⎜⎜⎝
⎛=⎟⎟
⎠
⎞⎜⎜⎝
⎛=h
x(i, j)Vertical ISI Horizontal ISI Channel
Detector
r(i, j)
w(i,j)
y(i, j)
Separable Channel Response
),( jix∧
Serial Turbo Codes
Rate 1/2 Systematic Encoder Rate 1/1 Non-Systematic Encoder
Separable 2-D ISI
x11 x12
x22x21
x13
x23
x14
x24
x34x33x32x31
x41 x42 x43 x44
-1 -1 -1 -1-1
-1
-1
-1 -1 -1 -1-1-1
-1-1-1-1-1
-1-1
x21 x22 x23 x24
Detection Performance
-5
-4
-3
-2
-1
5 6 7 8 9 10 11 12SNR (dB)
Bit-
Erro
r-R
ate
In lo
g10
ZF_Horizontal_MAP
Vertical_Horizontal_MAP
MMSE_Equalization
Iterative Decoder Diagram I
Equalization MAPDetector
LDPCDecoder
r(i, j)
Row-by-Row Detector
Conventional One-Dimensional Iterative Decoder
Extrinsic Information
Hard Decision
Iterative Decoder Diagram II
VerticalDecoder
HorizontalDetector
LDPCDecoder
r(i, j)
Passing TransitionProbability
Conventional Iterative Detector
Extrinsic Information
Hard Decision
Decoding Performance
Performance of Iterative Decoding Diagram I and II
-6
-5
-4
-3
-2
-1
0
0 1 2 3 4 5 6SNR (dB)
Bit-
Erro
r-Rat
e In
log1
0
ISI-free
Diagram I_Iter1
Diagram I_Iter2
Diagram I_Iter3
Diagram II_Iter1
Diagram II_Iter2
Diagram II_Iter5
Iterative Decoder Diagram III
VerticalMAP Detector
HorizontalMAP Detector
LDPCDecoder
Z
⊕
⊕
⊕
⊕-
-
-
-+
+
+
+
D
)( , jixL−
)( , jiyL−
)( ,|
jiyL
)( , jiLDPC xL
Decoding Performance
Performance of Iterative Decoding Diagram III
-6
-5
-4
-3
-2
-1
0
0 1 2 3 4 5SNR (dB)
Bit-
Erro
r_R
ate
in lo
g10
Diagram III_Iter1_SubIter2
Diagram III_Iter2_SubIter2
Diagram III_Iter3_SubIter2
Diagram III_Iter4_SubIter2
Diagram III_Iter10_SubIter2
ISI_free
Performance Comparison
-6
-5
-4
-3
-2
-1
0
0 1 2 3 4 5 6
SNR (dB)
Bit-
Erro
r-R
ate
In lo
g10
ISI-freeDiagram III_Iter10_SubIter2Diagram II_Iter5Diagram I_ZF_Equa_Iter3Diagram I_M M SE_Equa_Iter10
Conclusions and Discussion• MMSE equalization and decoding
• good performance with iterative equalization• low complexity FIR implementation
• Message passing algorithms• full-graph algorithm performance deteriorated due to
short cycles• ordered subsets message passing gives best
performance for general 2-D ISI• complexity proportional to code block-length
• Separable ISI decoding• best performance for separable 2-D ISI• low complexity . . .• approximate channel response by separable response