Joint Detection and Decoding
Schemes for Two Dimensional
Data Storage Systems
Joseph A. O’Sullivan
Naveen Singla
Joint work with Y. Wu and R. S. Indeck
Electronics Systems and Signals Research Laboratory
Magnetic Information Science Center
Washington University
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!
Problem Description
• Reliable data retrieval from channels having
2-D ISI
Advanced storage media: Patterned media
As bit aspect ratio reduces inter-track
interference becomes significant
Optical memories
Conventional PRML System
Receive
Filter
(low pass)
Adaptive
Filter
Viterbi
Detector
Magnetic
Recording
Channel
Timing &
Tap Updating
Wide-Band
Noise
nT+
e
ny nxnxny
,....3,2,1,)1)(1()( nDDDP n
No generalization to 2D
Joint Equalization and Decoding
Schemes
• General 2D ISI
Using 2D MMSE equalization and decoding
Using novel message-passing algorithms that
take advantage of the 2D dependence
• Separable 2D ISI
Using turbo equalization
Singla et al., “Iterative decoding and equalization for 2-D recording channels,” IEEE Trans.
Magn., Sept. 2002.
Channel Model
• x(i,j) {+1,-1}
• Channel ISI is 2D and linear
• Noise assumed to be AWGN
LDPC
EncoderEqualizer
LDPC
Decoder
Channel
ISI
Channel
),( jia ),( jix ),( jir ),( jix ),( jia
),( jiw
2D Intersymbol Interference
1111
11
11
11
1111
21
21
11211
kkkk
k
xxx
x
xxx
1111110
110
12
1111110
100100
kkkkkk
kkkkkk
k
kk
k
rrrr
rrrr
r
rrrr
rrr
GUARD BAND
),(),(),(),(2
0,
jiwnmhnjmixjirnm
),( jiw
cbc
bab
cbc
h
MMSE Equalization and
Decoding
• Equalizer designed assuming inputs to be
Gaussian
LDPC
EncoderEqualizer
LDPC
Decoder
Channel
ISI
),( jia ),( jix ),( jir ),( jix ),( jia
),( jiw
Performance
MMSE Equalization and Decoding
-7
-6
-5
-4
-3
-2
-1
0
0 1 2 3 4 5 6 7
SNR (dB)
Bit
Err
or
Ra
te,
Lo
g B
as
e 1
0
BIAWGN Capacity
LDPC No ISI
Wiener
220.0366.0220.0
366.0819.0366.0
220.0366.0220.0
h
Iterative MMSE Equalization and
Decoding
• Soft information, estimated mean of the codeword,
passed from LDPC decoder to equalizer
]][***[*][
)1Pr()1Pr(][
xEhrWxEx
xxxE
Extrinsic Information
LDPC
EncoderEqualizer
LDPC
Decoder
Channel
ISI
),( jia ),( jix ),( jir ),( jix ),( jia
),( jiw
Performance
Iterative MMSE Equalization and Decoding
-7
-6
-5
-4
-3
-2
-1
0
0 1 2 3 4 5 6 7
SNR (dB)
Bit
Err
or
Ra
te,
Lo
g B
as
e 1
0
BIAWGN Capacity
LDPC No ISI
Itrw iener_10
Wiener
220.0366.0220.0
366.0819.0366.0
220.0366.0220.0
h
Full Graph Message-Passing
Check Nodes
Codeword Bit Nodes
Data Nodes
jijijijijiji wxxxxr ,1,11,,1,, 25.05.05.025.05.0
5.01h
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)
From
Check
Nodes
Performance
Full Graph Message-Passing
-7
-6
-5
-4
-3
-2
-1
0
0 1 2 3 4 5 6 7
SNR (dB)
Bit
Err
or
Ra
te,
Lo
g B
as
e 1
0
BIAWGN Capacity
LDPC No ISI
Full graph_50
Itrw iener_10
Wiener
220.0366.0220.0
366.0819.0366.0
220.0366.0220.0
h
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)
From
Check
Nodes
Kschischang et al., “Factor graphs and the sum-product algorithm,” IEEE Trans. Inform.
Theory, Feb. 2001.
Ordered Subsets Message-Passing
• From Imaging – Data set is grouped into
subsets to increase rate of convergence
• For Decoding – Observed data is grouped
into subsets to eliminate short length cycles
in the Channel ISI graph
H. M. Hudson, and R. S. Larkin, “Accelerated image reconstruction using ordered subsets
of projection data,” IEEE Trans. Medical Imaging, Dec. 1994
Grouped ISI Graph
• Labeling of data nodes into 4 subsets
• For each iteration use data nodes of one label only
J. A. O’Sullivan, and N. Singla, “Ordered subsets message-passing,” Submitted to Int’l
Symp. Inform. Theory 2003.
Performance
Ordered Subsets Message-Passing
-7
-6
-5
-4
-3
-2
-1
0
0 1 2 3 4 5 6 7
SNR (dB)
Bit
Err
or
Ra
te,
Lo
g B
as
e 1
0
BIAWGN Capacity
LDPC No ISI
Ord Subsets_100
Full graph_50
Itrw iener_10
Wiener
220.0366.0220.0
366.0819.0366.0
220.0366.0220.0
h
Performance
Joint Equalization and Decoding Schemes
-7
-6
-5
-4
-3
-2
-1
0
0 0.5 1 1.5 2 2.5
SNR (dB)
Bit
Err
or
Ra
te,
Lo
g B
as
e 1
0
BIAWGN Capacity
LDPC No ISI
Ord Subsets_100
Full graph_50
Itrw iener_10
Wiener
019.0098.0019.0
098.0980.0098.0
019.0098.0019.0
h
Performance
Joint Equalization and Decoding Schemes
-7
-6
-5
-4
-3
-2
-1
0
0 1 2 3 4 5 6 7
SNR (dB)
Bit
Err
or
Ra
te,
Lo
g B
as
e 1
0
BIAWGN Capacity
LDPC No ISI
Ord Subsets_100
Full graph_50
Itrw iener_10
Wiener
0353.00
353.0707.0353.0
0353.00
h
Performance
Joint Equalization and Decoding Schemes
-7
-6
-5
-4
-3
-2
-1
0
0 0.5 1 1.5 2 2.5
SNR (dB)
Bit
Err
or
Ra
te,
Lo
g B
as
e 1
0
BIAWGN Capacity
LDPC No ISI
Ord Subsets_100
Full graph_50
Itrw iener_10
Wiener
0098.00
098.0981.0098.0
0098.00
h
Consider Separable 2D ISI
Wu et al., “Iterative detection and decoding for separable two-dimensional intersymbol
interference” Submitted to IEEE Trans. Magn., June. 2002.
A Separable 2D ISI
• Advantages of Separable 2D ISIApply existing one-dimensional equalization methodsReduced Detector Complexity
5.015.0
1
25.05.0
5.01h
x(i, j)Row ISI Column ISI
Channel
Detector
r(i, j)
w(i,j)
y(i, j)
Separable Channel Response
Tüchler et al., “Turbo equalization: principles and new results,” IEEE Trans. On Comm.,
May 2002.
Row-by-Row Decoder
• MMSE and Zero-forcing criteria used for Equalization
Column
Equalization
MAP
Detector
LDPC
Decoder
r(i, j)
Conventional One-Dimensional Iterative Decoder
Extrinsic Information
Hard Decision
Performance
Row-by-Row Decoder
-6
-5
-4
-3
-2
-1
0
0 1 2 3 4 5
SNR (dB)
Bit-E
rro
r-R
ate
In
lo
g1
0
LDPC No ISI
Row -by-Row : MMSE_iter10
Row -by-Row : ZF_iter10
25.05.0
5.01h
Row-and-Column Decoder
• Inputs to column detector are not binary
Column MAP
Detector
Row MAP
Detector
LDPC
Decoder
r(i, j)
Conventional One-Dimensional Iterative Detector
Extrinsic Information
Hard Decision
Performance
Row-and-Column Decoder
-6
-5
-4
-3
-2
-1
0
1 2 3 4 5
SNR (dB)
Bit
Err
or
Rate
, Log B
ase 1
0
LDPC No ISI
Row -and-Column_10
Ordered Subsets_100
Full Graph_50Itrw iener_10
Wiener
Row -by-Row : ZF_iter10
25.05.0
5.01h
Conclusions
• MMSE equalization and decoding
Good Performance with Iterative Equalization
• Message-passing algorithms
Full graph algorithm performance deteriorated due to short cycles
Ordered subsets message-passing gives best performance for general 2D ISI
• Separable ISI decoding
Best performance for separable 2D ISI
Low complexity
Approximate channel response by separable response
Performance
Ignoring ISI
-6
-5
-4
-3
-2
-1
0
0 2 4 6 8 10 12 14
SNR [dB]
Bit E
rror R
ate
in log10
ISI-free
ISI ignored
Block length 10000 regular (3,6) LDPC code
MMSE Equalization and Decoding
-6
-5
-4
-3
-2
-1
0
0 2 4 6 8 10 12 14
SNR [dB]
Bit E
rro
r R
ate
in
lo
g1
0
ISI-free
Wiener
ISI ignored
Performance
Iterative MMSE and Decoding
-6
-5
-4
-3
-2
-1
0
0 1 2 3 4 5
SNR [dB]
Bit E
rror
Rate
in log10
ISI-free
Itr Wiener_10
Wiener
Performance
Full Graph Message-Passing
-6
-5
-4
-3
-2
-1
0
0 1 2 3 4 5
SNR [dB]
Bit E
rro
r R
ate
in
lo
g1
0
ISI-free
Full Graph_50
Itr Wiener_10
Wiener
Performance
Ordered Subsets Message-Passing
-6
-5
-4
-3
-2
-1
0
0 1 2 3 4 5
SNR [dB]
Bit E
rro
r R
ate
in
lo
g1
0
ISI-free
Ordered Subsets_100
Full Graph_50
Itr Wiener_10
Wiener
Performance