相変化型光ディスクへの多値記録技術

相変化型光ディスクへの多値記録技術
Multi-level Recording on Phase-change Optical Discs
清水 明彦*
阪上 弘文*
門川 雄一*
前川 博史*
竹内 弘司*
Akihiko SHIMIZU
Kohbun SAKAGAMI
Yuichi KADOKAWA
Hiroshi MAEKAWA
Koji TAKEUCHI
田代 浩子
*
Hiroko TASHIRO
高津 和典
**
Kazunori TAKATSU
要
旨
短波長化と高NA ( Numerical Aperture：開口数) 技術の導入により，CD-R/RWから書き換え型
DVDへと光ディスクの大容量化が進む一方，多値記録による高密度，高転送レートの実現が注目
されており，特に多値記録は，光ディスク装置の光学系を変更せずに容量を増加できるので，容
易に商品の付加価値を高められる利点がある．相変化型光ディスクであるDVD+RWディスクを用
いて，多値データの値によって記録マーク長を変化させる面積変調により，8値記録を実現した．
信号再生時の符号間干渉を隣接データとの相関と見なし，パタン認識技術によって多値データを
判定する多値判定方式 Data Detection using Pattern Recognition ( DDPR)及び，8値データ（3ビット）
の最下位ビットに制限を加えたデータ変調方式 LSB (Least Significant Bit) Limited Modulation (LLM)
を新規に開発し，両者の組合せにより，DVDの1.7倍容量を達成した．
ABSTRACT
The capacity of optical discs has been increased by using laser diodes with shortened wavelength and
lenses with a larger numerical aperture. This has enabled, for example, the development of rewritable
DVDs with greater disc capacity than CD-R/RWs. Recently, however, an optical disc system using
multi-level recording has been recognized as a more effective means of achieving higher data capacity and
transfer rates. Because this system requires no changes to the pick-up head in the disc drive, it makes it
easy to increase the values of the disc system. Following up on this, in this paper the multi-level
recording on a rewitable DVD is reported. The eight-level data are recorded on a DVD+RW disc, i.e., a
phase-change rewritable optical disc, using area modulation where the length of the recording mark on
the disc varies with the level of data. The newly-developed technologies that make this possible are Data
Detection using Pattern Recognition (DDPR) and LSB (Least Significant Bit) Limited Modulation (LLM).
DDPR detects multi-level data by regarding inter-symbol interference as the correlation between
adjacent data. LLM modulates binary data into eight-level (three-bit) data where the LSB is bound by
certain rules. Coupling these technologies makes it possible to achieve a DVD with capacity 1.7 times
that of a conventional DVD.
*
**
研究開発本部 光メモリー研究所
Optical Memory R&D Center, Research and Development Group
画像システム事業本部 ソフトウエア研究所
Software Research Center, Image System Business Group
Ricoh Technical Report No.28
34
DECEMBER, 2002
cell.
1．INTRODUCTION
A rewritable phase change optical disc that is compatible with
the DVD-ROM was released as the next generation of CD-RW.
The data capacity of this rewritable DVD disc is 4.7G bytes and
that is about seven times that of a CD-RW. The data transfer rate
is about 88 Mbps for reading and about 26 Mbps for writing.
However, greater data capacity is needed to cope with the
increased amount of information that has been accumulated over
the last few years. And a higher data transfer rate will also be
needed to keep pace with the speedups of computer systems. The
current aim of optical disc development is to increase both data
Fig.1
The conventional method, the relationship between
mark and output signal in bi-level recording.
Fig.2
The multi-level recording method, the relationship
between mark and output signal for eight-level
recording.
capacity and the data transfer rate. One of the solutions to this
task is focused to use multi-level recording without any changes
being made to the optical pick up head. [1]
We have developed two original methods for multi-level data
recording on optical discs. One is a data detection method named
Data Detection using Pattern Recognition (DDPR) [2] in order to
enlarge the recording margin and the playback margin that are the
parameter defined for recording quality. The other is a new data
modulation
process,
LSB
(Least
Significant
Bit)
Limited
Modulation (LLM), in corporation with DDPR in order to reduce
data errors. In this paper we have explained DDPR and LLM, in
addition to the experimental results that the coupling LLM with
DDPR was effective in reducing data errors using DVD+RW disc
system.
2-2
Sigma to Dynamic Range
An evaluation of data recording quality involves the time
2．MULTI-LEVEL RECORDING
variations of playback signal's edges (jitter) in the conventional
case of bi-level recording. In the case of multi-level recording,
2-1
Comparison of recording and playback
Sigma to Dynamic Range (SDR) has been proposed for the
methods
evaluation of recording quality. Fig.3 shows the definition of SDR.
Fig.1 and Fig.2 show the comparison between the conventional
“SDR is the ratio of the standard deviation of a recorded level to
method and multi-level recording method. In the conventional
the total dynamic range between minimum and maximum
method, information is recorded and reproduced as the time
reflectivity.” Usually the bit error rate (BER) needs to be about
interval data. The data is equivalent to the mark length or the
10-5 before the performance of the error correction that is
space length. On the other hand, in the multi-level recording,
generally used for optical discs. The SDR needs to be less than
information is recorded and reproduced as the signal amplitude.
2.0% for data detection using a fixed threshold level to achieve a
The data is equivalent to the rate of mark size occupying in a data
BER of 10-5. [3] To attain stability in the tolerance of compatibility
Ricoh Technical Report No.28
35
DECEMBER, 2002
controlled so that the reflectivity from the mark would have eight
and mass production, it is preferable to enlarge the SDR margin.
different levels. The higher the recording density becomes, the
smaller the data cell becomes in comparison with the diameter of
beam spot, as shown in Fig.5. Inter-symbol interference (ISI)
occurs and varies the amplitude of the playback signal in the
center of data cell.
Fig.3
2-3
SDR has been proposed for the evaluation of
recording quality. SDR is calculated as a ratio of the
averaged value of σ in all levels to DR.
Multi-level Recording Marks
Fig.5
Fig.4 shows the TEM image of multi-level recording marks on
Explanation of the inter-symbol interference (ISI).
Premark position is n-1, postmark position is n+1.
There are three data cells in a beam spot.
the phase-change disc of DVD+RW. The width of a data cell is
0.74 μm and its length is 0.4μm. Under this geometry condition
Fig.6 shows an example of amplitude variations caused by ISI.
the maximum mark of multi-level recording is equivalent to the
Playback signal waveforms corresponding to three continuous
minimum mark of DVD, and the recording density becomes double
multi-level data (n-1, n, n+1) are shown. T(i, j, k) represents the
of DVD’s. It needs to control the write pulse signal in order to
amplitude of multi-level data(n), whose level is j , between
record these small marks accurately.
predata(n-1), whose level is i, and postdata(n+1), whose level is k.
Though T(0,1,0) and T(7,0,0) are very close to each other, they
must be detected as different levels(1 and 0). It is difficult to
distinguish between them for data detection using a fixed threshold
level.
Fig.4
TEM image of multi-level recording marks, cell size is
0.74µm ×0.40µm, Track pitch is 0.74µm, mark
length is 0 to 0.36µm.
3．DDPR
3-1
Fig.6
Principle of DDPR
Example of ambiguous waveforms. Premark position
is n-1 and postmark position is n+1. The signal value
of T(7, 0, 0) is close to that of T(0, 1, 0).
Multi-level data is recorded as a mark of variable length in a
The ISI still remains after equalization using a linear Finite
data cell on a spiral track of a disc. The length of the mark was
Impulse Response (FIR) filter. This ISI is caused by pre and post
Ricoh Technical Report No.28
36
DECEMBER, 2002
data and its effect seems to be nonlinear. The effect of ISI
found in Fig.7. The equalized signal is pre-detected using fixed
becomes larger when the number of data levels is increased and
threshold levels.
the dynamic ranges of each level are narrowed. Therefore, we
developed DDPR that detects multi-level data using a correlation
obtained from three continuous data.
3-2
Signal Process of DDPR
DDPR is composed of two steps. One is the process of “making
pattern table” which learn the ISI and the variation of signal
amplitude. The other is the process of “data detection using the
pattern table”.
Fig.7 shows the signal processing flow for making a pattern
Fig.8
table before using data detection. The test signal read from the
disc is converted into a digital signal using an A/D converter. The
Data detection using the pattern table. The
comparator outputs a candidate that has minimum
error as a detected data.
digital signal is equalized to remove inter-symbol inference using a
five-tap equalizer. The output signal of the equalizer is averaged
Fig.9 shows the amplitude histograms of eight-level data after
using the same known data. The averaged value is stored into the
equalizing the test signal. The threshold levels are set between the
table. T(i, j, k) presents the averaged value of the equalized signal,
distributions of each level. Pre-detection fixes the levels (i and k)
corresponding to the center data (j:0 to 7) between pre and post
of pre and postdata using the threshold levels. After the
data (i, k: 0 to 7).
pre-detection, eight candidates (T(i, j, k): j is 0 to 7.) are picked
up from the pattern table. The errors between the equalized signal
and the candidates are calculated. The comparator outputs a
candidate that has minimum error as a detected signal. This
pattern recognition process was based on ISI from pre/post marks
after the equalization.
Fig.7
Making pattern table. T(i, j, k) presents the averaged
value of the equalized signal corresponding to the
center data (j:0 to 7) between pre and postdata (i, k:
0 to 7).
Fig.9
3-3
Fig.8 shows the signal processing flow of the data detection
The amplitude histograms of eight-level data after
equalizing. Cell length is 0.40 µm. SDR is 2.5%.
Experiment and Results of DDPR
using the pattern table generated above. The data signal is
Multi-level data was recorded on a DVD+RW disc of track
converted into a digital signal and equalized using the same flow
pitch 0.74 μm at linear velocities, ranging from 3.5 to 7.0 m/s
Ricoh Technical Report No.28
37
DECEMBER, 2002
using a DVD+RW drive. A laser diode of 650nm wavelength and an
4．LLM
objective lens of 0.65-NA (Numerical Aperture) are mounted on
its pick-up head. The effect of the DDPR on BER and SDR for
4-1
various data-cell lengths was investigated. For example, in Fig.9,
Principle of LLM
the averaged SDR was 2.5%, where the data-cell length was about
Fig.11 shows the example of detected level deviation after
0.40 μm and the data capacity was twice that of the DVD. Fig.10
DDPR process. We confirmed that the level error is within ±1
shows that using DDPR reduces BER more effectively than such
from this result. This deviation of ±1 level causes the bit reverse
data detection using fixed threshold levels as pre-detection on the
of LSB. Therefor it’s possible to correct the level error by
-5
BER vs. SDR plane. The BER of the order of 10 was achieved
restricting LSB of multi-level data.
using DDPR for SDR ≦2.7%. DDPR enlarged a SDR margin from
2.0 to 2.7%. DDPR using code modulation should improve BER
and should be able to design for a feasible multi-level recording
system.
Fig.11 Example of level deviation histogram after DDPR
process, the level error is ±1.
Fig.10 Comparison between DDPR and Pre-detection,
DDPR enlarged SDR margin from 2.0% to 2.7%.
Ricoh Technical Report No.28
38
DECEMBER, 2002
patterns where two or more bits are different from each other. For
4-2
Modulation using LLM
example, (D0, D1, D2) = (0, 0, 0), (0, 0, 1), and (0, 1, 0) are
Fig.12 shows data arrangements before and after LLM. The
converted into (L0, L1, L2, L3) = (1, 1, 1, 1), (1, 1, 0, 0), and (1,
11-bit input data (D0, D1, … D10) are converted into a set of
0, 1, 0), respectively and where P = 0. Comparing (1, 1, 1, 1) with
four symbols (S0, S1, S2, S3). As a symbol is eight-level data, it
(1, 1, 0, 0) and (1, 1, 0, 0) with (1, 0, 1, 0), there are two different
contains three bits. The LSBs (L0, L1, L2, L3) of the symbols are
bits in each pattern. One redundant bit enlarges the code distance
generated from three bits (D0, D1, D2) in the input data using a
(Hamming distance). Moreover, P generates a correlation between
conversion table. The rest of the input data (eight bits: D3, D4,
sets. This correlation is also useful in the demodulation process
… D10) are arranged directly into the MSB (Most Significant Bit)
similar to maximum likelihood decoding.
side of the symbols sets as shown in Fig.12.
4-3
Fig.13 shows the conversion table used in LLM. The three bits
Demodulation using LLM
(D0, D1, D2) are converted into four bits (L0, L1, L2, L3). This
Fig.14 shows the signal processing flow of LSB Limited
table has two conversion rules that are selected by a parameter
demodulation. An input multi-level signal is detected to generate
“P” (0 or 1) defined in every set. P (n) in the n-th set is defined as
two candidates for a symbol by using threshold levels. Because the
the result of exclusive-or operation between P (n-1) and L3 (n-1)
conversion table limits the LSB of a symbol, a symbol candidate
in the previous (n-1)-th set. The initial value of P (1) is 0. The
should be an even or odd level. The nearest two (even and odd)
four-bit converted data pattern in the table differs from adjacent
levels are chosen as candidates from eight levels. Using two
Input data (11 bits)
D0 D1 D2 D3 D4 D5 D6 D7 D8 D9 D10
3 bits
Conversion
table
Set of modulated
data (12 bits)
S0
D3
D4
L0
8 bits
4 bits
S1
D5
D6
L1
S2 S3
D7 D9 MSB
D8 D10
L2 L3 LSB
Fig.12 Data arrangements before and after LLM, the 11-bit input data (D0, D1, … D10) are converted into a set of four
symbols (S0, S1, S2, S3). The LSBs (L0, L1, L2, L3) of the symbols are generated from three bits (D0, D1, D2).
Fig.13 The conversion table used in LLM. The three bits (D0, D1, D2) are converted into four bits (L0, L1, L2, L3). This table
has two conversion rules that are selected by a parameter "P" (0 or 1) defined in every set.
Ricoh Technical Report No.28
39
DECEMBER, 2002
Input signal
generate candidates for a symbol
generate candidates for a set
Threshold levels
P, Conversion table
generate reference signal for candidates
accumulate signal errors
search the least error
Detected data
Fig.14 The signal processing flow of LSB Limited demodulation.
Fig.15 Signal Process of data detection coupling LLM with DDPR
candidates for a symbol, eight candidates for foursymbols in a set
correlation between sets generated by P. If P is incorrect, symbol
are generated according to the parameter P in the set and the
errors may propagate in several sets. The DDPR is able to
conversion table. A reference signal value corresponding to a
terminate the error’s propagation. Thus, the DDPR and LLM
candidate for a symbol is generated. A signal error (difference)
compensate for each other’s faults
between the input signal and the reference signal value for each
4-5
symbol is accumulated in several sets. The candidate with the
Experiment and Results of LLM + DDPR
Fig.16 is a block diagram of the experimental system. Random
least error is output as detected data.
data was generated and LL modulated into multi-level data using a
4-4
Coupling LLM with DDPR
personal computer. A write-pulse generator converted the
Fig.15 shows the signal processing flow of data detection
multi-level data into a driving signal for a laser diode in an optical
coupling LLM with DDPR. The playback signal read from an
pick-up head (PU). Multi-level data was recorded on a DVD+RW
optical disc is converted into a digital signal using an A/D
disc the same DVD+RW drive as mentioned in 3-3. The playback
converter. The digital signal is equalized to remove inter-symbol
signal was converted into a digital signal using an A/D converter
interference using a 7-tap equalizer. After equalization DDPR and
with a clock synchronized with the playback signal using a PLL
LL demodulation are processed in parallel. The LSBs of output
(Phase-Locked Loop) circuit. The data detection shown in Fig.15
data in a set from DDPR are checked as to whether they obey the
was processed using software on the personal computer.
conversion rule of the table used in LLM. When they match the
Fig.17 shows an example of playback signal and clock signal
rule, the result of DDPR is output a detected data through a
waveforms. Using DDPR and LLM together achieved a BER (Bit
selector. If they do not match, the result of LL demodulation is
Error Rate) of the order of 10-5 before error correction where
output as detected data. The DDPR detects multi-level data using
recording density was 1.7 times that of a conventional DVD,
the correlation between pre- and post data. The LSBs in a set
assuming that a format efficiency of ECC (Error Correction Code)
can be used as error detection code. The LL demodulation detects
was the same as that of a standard DVD. The BER was 10-4 where
multi-level data according to the conversion rule in LLM and the
recording density was twice as great. Using only DDPR, the
Ricoh Technical Report No.28
40
DECEMBER, 2002
recording density was 1.4 times where the BER was 10-5..
Fig.16 The block diagram of the experimental system.
Fig.17 Example of playback signal and clock signal waves.
5．CONCLUSIONS
We developed the new method of the coupling LLM with DDPR
for the multilevel recording. Our experiments confirmed that
coupling LLM with DDPR was effective in reducing data errors and
increasing recording density. This effective performance enabled
to make 1.7 times recording density of a conventional DVD. We
intend to refine this method and apply it to a blue-laser optical
disc system.
REFERENCES
[1]
M. P. O’Neill et al.: "Over 20GB Rewritable Multi-level Recording
using Blue Laser and Growth-dominant Phase-change Optical Discs",
Technical Digest of ISOM 2000, Chitose, Japan, (September 2000),
pp. 234-235.
[2]
A. Shimizu et al.: “Data Detection using Pattern Recognition for
Multi-level Optical Recording”, Technical Digest of ISOM 2001,
Taipei, Taiwan, (October 2001), pp. 300-301.
[3]
S. McLaughlin et al.: "Advanced Coding and Signal Processing for
Multilevel Write-Once and Rewritable Optical Storage", Technical
Digest of ODS 2001, Santa Fe, New Mexico, (April 2001), pp. 76-78.
Ricoh Technical Report No.28
41
DECEMBER, 2002