501 KKU Res J 13 (4) : May 2008
Abstract
In this paper, the effect of cutoff frequency of the front-end low-pass filter on the performance of
perpendicular recording channel is investigated. With positive and negative deviation from the nominal value,
the partial response maximumlikelihood (PRML) system shows more degradation by the positive deviation
than the negative one. The noise-predictive maximumlikelihood (NPML) system shows the results otherwise.
rrr--: or ca:orr r:-,a-a-, or to,ass ri|:-: oa :|-
r-:ro:aa-- or r-:,-aai-a|a: Ma,a-:i- k--o:aia, c|aaa-|
Yupin Suppakhun
1
, Pornchai Supnithi
2
Faculty of Engineering, Research Center for Telecommunications and Information Technology(ReCCIT) and Research Center in Data Storage
Technology and Applications (D*STAR) King Mongkuts Institite of Tecnology Ladkrabang, Bangkok, Thailand,
*corresponding author; e-mail: s9060001@[Link]
1
, ksupornc@[Link]
2
Introduction
Current magnetic recording system employs
perpendicular media , thus the transition response is
vastly different from the longitudinal media with
existent dc component. The front-end components
in the read channel chip consists of variable gain
amplifier (VGA), amplitude asymmetry correction,
thermal asperity (TA) detection and correction and
low-pass filter. The low-pass filter has a function
of suppressing the out-of-band noise. Its coefficients,
cut off frequency, frequency boost factor among
others can be adapted for each media and read-write
system so that bit error rate (BER) can be
optimized. In this paper, we aim to investigate the
influence of the cutoff frequency deviation to the
system performance, in particular, the PRML and
NPML detector in the environment of electronics
noise and jitter noise. In Section 2, we overview the
readback system model, the NPML system and
related parameters. The simulation results and
discussions are then illustrated in Section 3. The
conclusion is described in Section 4 and the
references are in Section 5.
Readback System Model
The readback system block diagram for the
perpendicular magnetic recording is shown in
Figure 1.
Figure 1. The readback system block diagram
Presented in 1
st
Data Storage Technology Conference (DST-CON 2008)
Effect of Cutoff Frequency of Lowpass Filter on the
Performance of Perpendicular Magnetic Recording Channel
502 . 13 (4) : 2551
The binary random sequences a
k
{1} are
input of the channel, where k represents the discrete
time index, k =1K (K is total number of transmit-
ted bits). A data input sequence with bit period T is
filtered by ideal differentiator (1-D) to form a tran-
sition sequence b
k
{-2,0,2}, where b
k
=
{2}
corresponds to a positive and negative transition, and
b
k
= {0} corresponds to the absence of transition.
The sequence b
k
passing through the channel is
convolved with the transition response.
The transition response for perpendicular
recording can be written as(Kovintavewat et al.,
2002).
2t ln2
g(t)= erf
PW50
(1)
where erf (.) is an error function defined as
2
-
0
2
( )
x
z
erf x e dz
=
, and PW50 determines the
pulse width of the derivative of g(t). The recorded
normalized density is defined by 50/ ND PW T = ,
hence, the readback signal r(t) can be expressed as
( ) ( ) ( )
k
k
r t b g t kT tk n t
=
= + +
(4)
where n(t) is additive white Gaussian noise
(AWGN) with two-sided power spectral density
No/2, and t
k
is jitter noise modeled as a random
shift in the transition position, which has a Gaussian
distribution function with zero mean and variance
j
2
,where
j
isspecified as a percentage of T and
t
k
/ 2.
The readback signal r(t) is filtered by a
Butterworth lowpass filter with cutoff frequency at
1/2T, which is sampled at a symbol rate. Its function
is to eliminate out-off-band noise.
The detection process is composed of two
components. The first component is a noise predictive
filter that reduces distortion(noise) from equalized
signal. The second component is the Viterbi detector
based on trellis of the PR target with adjusted trellis
adjusted by output of the noise predictor.
Noise-predictive maximum likelihood
(NPML) detector
Let y
k
be output data sequence of the PR
equalizer at instant k. The finite impulse
response(FIR) filter has the polynomial of the PR
target in the form of F(D)=(1 + f
1
D + f
2
D
2
+...+f
N
D
N
), where the fi(i = 2,...,N) is the
coefficients of the filter. The equalized output is
1
n
k k i k i k
i
y a f a w
=
= + +
(3)
where w
k
is the colored noise sequences at
the output of equalizer. The power of the colored
noise component can be reduced by noise prediction.
The NPML system uses a predictor with N-coeffi-
cients. Given the transfer polynomial of the FIR noise
predictor filter is P(D)=(1 + p
1
D + p
2
D
2
+...+p
N
D
N
) or, equivalently, E (D) =[1-P(D)]
denotes the transfer polynomial of the predictor
error filter, then the whitened noise component e
k
from the predictor can be computed by
1
,
k k k
N
k k i i
i
e w w
w w p
=
=
=
where the noise predicted sample
k
w can be defined
as
1
N
k k i i
i
w w p
=
=
(5)
503 Effect of Cutoff Frequency of Lowpass Filter on the
Performance of Perpendicular Magnetic Recording Channel
KKU Res J 13 (4) : May 2008
The coefficients of a noise predictor filter are
determined by solving the system of well-known
normal equation given by and variance
j
2
, where
j
is
1
( ) ( ) , 1, 2,...,
N
ww i ww
j
R i p R i j i N
=
= =
(6)
where R
ww
is autocorrelation function, which can be
written in the matrix form as
1
2
(1) (0) (1) ( 1)
(2) (1) (0) ( 2)
.
( ) ( 1) (1) (0)
ww ww ww ww
ww ww ww ww
ww ww ww ww N
R R R R N p
R R R R N p
R N R N R R p
=
r R
14243 1444444442444444443{
,
p
where R represents the square matrix, p
i
is
determined by
(7)
where p = [p
1
p
2
p
N
] and
r = [ R
ww
(1) R
ww
(2)...R
ww
(N)]
T
.
The NPML detection results from the
embedding the noise prediction/whitening process
into the branch metric computation of the Viterbi
detector. The output of noise predictor error filter Z
k
to viterbi detector in D domain term is
where H
ef f
(D) = (1 - g
1
D-g
2
D
2
-...-g
N
+
V
D
N+v
represents the transfer polynomial of effective target
which corresponds to noise predictor error filter,
where the gi (i = 1, 2,...,N) is the N-tap coefficients
of the effective target, v is the memory of PR target
and H(D) is partial response target, then the viterbi
detector uses a state trellis with the number of state
2
N+v
.
The branch metric of the NPML detector for
effective target samples corresponding to a transi-
tion from state p to state q takes the form
Z
k
= (y
k
)[1-P(D), (9)
and
H
ef f
(D) = ( H(D) ) [ 1- P(D) ] , (10)
2
( , ) ( , )
k k k
p q Z O p q =
where
k
(p, q) represents the branch metric
cost from state p to state q, and
k
O is noiseless
channel output from effective target (H
ef f
( D) )
defined as
*
k k eff
O a H =
(11)
(12)
Where * denotes the convolution operator.
Simulation Results and Discussions
In this section, we present BER simulation
results for various percentage of cutoff frequency
(%Wn) on infinite impulse response(IIR) low pass
filter, and investigate the BER performance in PRML
detector and NPML detector at DC-Attenuation target
(5 6 0 -1). In the simulations, the received
sequence S
k
is equalized by 21-tap finite impulse
response(FIR) filter calculated to minimize the mean-
square error (MMSE) of the equalizer output and
target response such that y
k
resembles d
k
. We process
each sector consisting of 4096 information bits and
let the parameter of normalized recording density
(ND) = 2.5, media jitter noise(J2) = 10%. and
noise predictive filter (NP_Tap) = 4 tap. The average
BER from the results are plotted versus the SNR(dB).
The percentages of cutoff frequency that deviates
from the nominal value 1/2T may be positive or
negative. For example, with +/-10% deviation, Wn
= 110 and Wn = 90, respectively , with +20%
Effect of Cutoff Frequency of Lowpass Filter on the
Performance of Perpendicular Magnetic Recording Channel
504 . 13 (4) : 2551
deviation, Wn = 120, and with -30% deviation,
Wn = 70. In Figure 2, the NPML detector at different
percentages cutoff frequency versus the system
performance is shown. We can see that the
performance degrades as %Wn changes, for example
at BER = 2 x 10
-4
with Wn = 70, there is an error
loss of 1.2 dB compared with the case when the
cutoff frequency is at 1/2T (Wn = 100).
Figure 2. NPML performance of percentages cutoff
frequency various base on DC-attenuation
target(5 6 0 -1)
Figure 3. PRML performance of percentages cutoff
frequency various base on DC-attenuation
target(5 6 0 -1)
In Figure 3 the BER performance of PRML
detector system affected by %Wn is shown. The
performance degrades more with positive deviation
from nominal cut frequency compared with the nega-
tive one. For example, at BER = 3x10
-3
with Wn =
120, there is an error loss of 0.6 dB compared with
the nominal cutoff frequency.
Conclusions
We have investigated the effect of cutoff
frequency percentage various of LPF in perpendicular
magnetic recording channel model in terms of the
BER performance of PRML and NPML detector for
DC-attenuation PR target. From the simulation
results, the NPML performance is affected by the
negative deviation of cutoff frequency more than the
positive one. On the contrary, the PRML perfor-
mance is affected by the positive deviation of cutoff
frequency more than the negative one.
References
Kovintavewat, P., Ozgunes, I., Kurtas, E., Barry, J.
R and McLaughlin, S. W. 2002.
Generalized Partial Response Target for
Perpendicular Recording with Jitter Noise,
IEEE Trans. on Magnetics, Vol. 38, Issue 5,
Sep. pp. 2340-2342.
Lee, J. and Lee, J. 2005. A Simplified
Noise-Predictive Partial Response Maximum
Likelihood Detection Using M-Algorithm for
Perpendicular Recording Channel, IEEE
Trans. on Magnetics, Vol. 41, Issue 2, Feb.
pp. 1064-1066.
Madden, M., Oberg, M., Wu, Z. and He, R. 2004.
Read Channel for Perpendicular Magnetic
Recording, IEEE Trans. on Magnetics, Vol.
40, No. 1, Jan., pp. 241-246.
"""""
