Post on 16-Jan-2016
slide 3UH, Manoa, December 2009
Disk drive
magnetic surface
track
magnify
slide 4UH, Manoa, December 2009
Magnetic recording channel
magnetization pattern on a track
media noise
media noise
Readback waveform
• Sample readback waveform• discrete time channel
• yk = xk - xk-1 + nk
• output: yk (real)
• noise: nk Gaussian, white, variance 2
slide 5UH, Manoa, December 2009
Prior results: 1-D channel
-10 -5 0 5 100
0.2
0.4
0.6
0.8
1
SNR [dB]
Cap
acit
y [b
its/
chan
nel
-use
]
water-filling
upper bound (Holsinger 1964)
upper bound
(Shamai et al. 1991)
lower bound
(Shamai et al. 1991
lower bound
(Shamai-Verdu 1992)
slide 6UH, Manoa, December 2009
New results: 1-D channel
-10 -5 0 5 100
0.2
0.4
0.6
0.8
1
SNR [dB]
Cap
acit
y [b
its/
chan
nel
-use
]
upper and lower bound almost
coincide
lower bound
(Kavcic 2001, Vontobel, Kavcic
2008)
upper bound
(Yang, Kavcic, Tatikonda 2005)
slide 7UH, Manoa, December 2009
LDPC codes: code/channel graph
C C C
V VV V
C
V V
c1 c4c3c2
s1 s3 s4 s6s5s2
T TTTT T
z1 z3 z4 z6z5z2
q0 q2 q3 q5q4q1 q6
slide 8UH, Manoa, December 2009
Noise tolerance thresholds
• Channel: 1-D
• Regular Gallager codes with variable node degree = 3
slide 9UH, Manoa, December 2009
Simulation results r = 0.5
slide 10UH, Manoa, December 2009
General 2-D Granular Media Model
• Granular medium: 2DMR (10Tb/sq in)
channel input
bits are written on grains
scan reading
channel output
granular medium
slide 11UH, Manoa, December 2009
Ordered statistics decoding on channels with memory
• Linear block codes (Reed-Solomon) are still in data storage standards (CDs, DVDs)
• Powerful codes, but difficult to decode on channels with memory
• We are developing ordered statistics techniques (pioneered by Fossorier and Lin) for channels with memory
slide 12UH, Manoa, December 2009
Summary
• Storage channels are channels with memory
• Research in– Channel modeling– Detection/estimation– Timing recovery– Information theory– Coding/Decoding