US20070096950A1  Structured set partitioning and multilevel coding for partial response channels  Google Patents
Structured set partitioning and multilevel coding for partial response channels Download PDFInfo
 Publication number
 US20070096950A1 US20070096950A1 US11/263,128 US26312805A US2007096950A1 US 20070096950 A1 US20070096950 A1 US 20070096950A1 US 26312805 A US26312805 A US 26312805A US 2007096950 A1 US2007096950 A1 US 2007096950A1
 Authority
 US
 United States
 Prior art keywords
 bits
 dc
 channel
 data
 free
 Prior art date
 Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
 Granted
Links
Images
Classifications

 G—PHYSICS
 G11—INFORMATION STORAGE
 G11B—INFORMATION STORAGE BASED ON RELATIVE MOVEMENT BETWEEN RECORD CARRIER AND TRANSDUCER
 G11B20/00—Signal processing not specific to the method of recording or reproducing; Circuits therefor
 G11B20/10—Digital recording or reproducing
 G11B20/10009—Improvement or modification of read or write signals

 G—PHYSICS
 G11—INFORMATION STORAGE
 G11B—INFORMATION STORAGE BASED ON RELATIVE MOVEMENT BETWEEN RECORD CARRIER AND TRANSDUCER
 G11B20/00—Signal processing not specific to the method of recording or reproducing; Circuits therefor
 G11B20/10—Digital recording or reproducing
 G11B20/10009—Improvement or modification of read or write signals
 G11B20/10046—Improvement or modification of read or write signals filtering or equalising, e.g. setting the tap weights of an FIR filter
 G11B20/10055—Improvement or modification of read or write signals filtering or equalising, e.g. setting the tap weights of an FIR filter using partial response filtering when writing the signal to the medium or reading it therefrom
Abstract
Description
 The present invention relates generally to coding and decoding data for transmission through a partial response channel, and more particularly but not by limitation to coding and decoding data for transmission through a data storage channel.
 Data communication channels generally include encoding of data before it passes through a communication medium, and decoding of data after it has passed through a communication medium. Data encoding and decoding is used, for example, in data storage devices for encoding data that is written on a storage medium and decoding data that is read from a storage medium. Encoding is applied in order to convert the data into a form that is compatible with the binary characteristic of the storage medium, and can include processes such as adding error correction codes, interleaving, turbo encoding, bandwidth limiting, amplification and many other known encoding processes. Decoding processes are generally inverse functions of the encoding processes. Encoding and decoding increases the reliability of the reproduced data.
 The increasing demand for high data rate communications systems and highdensity data storage devices has lead to intensive focus on implementation issues of encoding and decoding that provides a combination of low bit error rate (BER), high speed, low power coding and decoding, and low cost. In designing encoders and decoders, there are complex design tradeoffs that keep changing as technologies advance.
 As the performance in the area of higher speed and lower power consumption improve in other parts of a data storage device, there is a need to advance the design of the encoding and decoding in the data communication channel such that the encoding/decoding performance does not limit the overall performance of new design data storage devices.
 Embodiments of the present invention provide solutions to these and other problems, and offer other advantages over the prior art.
 Disclosed are an apparatus and a method of encoding and decoding user data. The encoding comprises receiving a block of user data and providing a block of DCfree encoded data. The encoding comprises performing a reverse rate1 mapping of the block of DCfree encoded data based on structured set partitioning to provides an intermediate data sequence. The encoding comprises generating redundant bits based on the intermediate data sequence using a multilevel encoder (MLC). The encoding comprises concatenating the redundant bits with the DCfree encoded data and communicating the result to a partial response channel.
 The decoding comprises detecting encoded data received from the binary medium and providing an estimated MLC encoded user data output. The decoding comprises decoding the MLC encoded data, and generating a MLC decoded data output. The decoding comprises DCfree decoding the MLC decoded data output to provide a decoded user data output.
 Other features and benefits that characterize embodiments of the present invention will be apparent upon reading the following detailed description and review of the associated drawings.

FIG. 1 is an isometric view of a disc drive. 
FIG. 1A illustrates an oblique view of a disc drive. 
FIG. 1B illustrates a first level set partition for a PR channel [1 11 12]. 
FIG. 2 illustrates a second level set partition for W_{0 }for a PR channel [1 11 12]. 
FIG. 3 illustrates a second level set partition for W_{1 }for a PR channel [1 11 12]. 
FIG. 4 illustrates third level set partition for W_{0,0 }of PR channel [1 11 12]. 
FIG. 5 illustrates a three level structured set partition for PR channel [1 11 12]. 
FIG. 6 illustrates a three level structured set partition with general multilevel coding. 
FIG. 7 illustrates multilevel encoding and decoding with structured set partitioning. 
FIG. 8 illustrates a SSP/MLC encoder with minimum disturbances to other constrained code encoding. 
FIG. 9 illustrates a decoder for SSP/MLC and RLL/DCF that corresponds with the encoder ofFIG. 8 . 
FIG. 10 illustrates a sequence detection for channel bits that are SSP/single parity encoded. 
FIG. 11 illustrates SSP decoding in which the first bit on which the parity check equation is effective is independent of the mth bit in the channel memory for most PR targets. 
FIG. 12 illustrates an SSP/MLC coded system with ECC and RLL/DCF encoding.  Disclosed here is a method and apparatus for channel coding useful for recording channel and other partial response channel applications. The proposed channel coding method is actualized via structured set partition (SSP) in conjunction with multilevel coding (MLC) and offers performance gains over conventional coding schemes with comparable complexity at the biterror rate (BER) level as well as the sector failure rate (SFR) level.
 The set partitioning and multilevel coding (MLC) jointly optimize modulation (signal mapping) and coding so that the code is optimized in Euclidean space rather than with traditional Hamming distances. Free Euclidean distances, rather than Hamming distances, define system performance, in particular at SNR regions that are of practical interest. For most set partition and MLC work on typical transmission channels, a multilevel/phase signal constellation is available. However, for magnetic recording channels, binary saturated recording is the only practical channel signaling method. In such binary saturated channels, realization of set partition and/or MLC is simulated by intentionally introduced intersymbol interference (ISI). Channel outputs from ISI channels can be conveniently characterized by a trellis, which in turn facilitates a set partition and/or MLC implementation.

FIG. 1A is an isometric view of a disc drive 100 in which embodiments of the present invention are useful. Disc drive 100 includes a housing with a base 102 and a top cover (not shown). Disc drive 100 further includes a disc pack 106, which is mounted on a spindle motor (not shown) by a disc clamp 108. Disc pack 106 includes a plurality of individual discs, which are mounted for corotation about central axis 109. Each disc surface has an associated disc head slider 110 which is mounted to disc drive 100 for communication with the disc surface. In the example shown inFIG. 1 , sliders 110 are supported by suspensions 112 which are in turn attached to track accessing arms 114 of an actuator 116. The actuator shown inFIG. 1 is of the type known as a rotary moving coil actuator and includes a voice coil motor (VCM), shown generally at 118. Voice coil motor 118 rotates actuator 116 with its attached heads 110 about a pivot shaft 120 to position heads 110 over a desired data track along an arcuate path 122 between a disc inner diameter 124 and a disc outer diameter 126. The heads 110 include a write head and a read head that couple to the desired data track which serves as a binary saturated communication medium. Voice coil motor 118 is driven by servo electronics 130 based on signals generated by heads 110 and a host computer (not shown).  The magnetic recording channel is approached as a partial response (PR) channel given by a polynomial f(D)=fo+f_{1}D+f_{2}D^{2}+ . . . +f_{m}D^{m }with integer (or real) coefficients. An output word y=(y_{1},y_{2}, . . . ,y_{N}) of the PR channel is defined as the linear convolution of the input coded (or uncoded) word x=(x_{1},x_{2}, . . . ,x_{N}) of length N with x_{1}=1 or −1, and the coefficients (f_{0},f_{1}, . . . ,f_{m}) as:
$\begin{array}{cc}{y}_{i}=\sum _{j=0}^{m}{f}_{j}{x}_{ij}& \mathrm{Equation}\text{\hspace{1em}}1\end{array}$
where i=1,2, . . . ,N, for i<1, and x_{i }is determined by the channel state. The relationship between the input and output words can be described by a trellis with 2^{m }states s∈{0,1}^{m}, where s_{i }is related to x_{i }by
x _{i}=2s _{i} −1. Equation 2  Given the state
S_{i}=(s_{i,0}, s_{i,1}, . . . , s_{i,m−1}) Equation 3  at time i=0,1, . . . ,N−1, the next state is defined as
S_{i+1}=(s_{i,1}, s_{i,2}, . . . , s_{i,m−1}, s_{i,m}). Equation 4  The label of the edge between S_{i }and S_{i+1 }is the channel output symbol y_{i }defined by Equation 1. The input and output relationship can be written in a more compact matrix form. Denote X^{−}=(x_{i−m}, . . . , x_{i−2}, x_{i−1}) and its binary version S^{−}=(s_{i−m}, . . . , x_{i−2}, s_{i−1}). The vector X^{−} and S^{−} are the states of the channel at time instant i. Given the trellis state S^{−} and a block of binary input S=(s_{i}, s_{i+1}, . . . , s_{i+L}) of length L, the output can be written as
Y=f(S ^{−} ⋄S)=X×F Equation 5  where X is the bipolar representation of vector S^{−} ⋄S, ⋄ is the concatenation operator, and F is an m+L by L matrix given by
$\begin{array}{cc}F=\uf603\begin{array}{cccc}{f}_{m}& 0& \dots & 0\\ {f}_{m1}& {f}_{m}& \dots & 0\\ \dots & \dots & \dots & \dots \\ {f}_{0}& {f}_{1}& \dots & {f}_{m}\\ 0& {f}_{0}& \dots & {f}_{m1}\\ \dots & \dots & \dots & \dots \\ 0& 0& \dots & {f}_{0}\end{array}\uf604.& \mathrm{Equation}\text{\hspace{1em}}6\end{array}$  The output corresponding to an input vector of length L and a channel state S^{−} (of length m) is a vector of length L. In the channel output sequence Y in an Ldimensional Euclidean space S, each input sequence S in conjunction with the channel state S^{−} correspond to a signal point in the Ldimensional space S. The mapping rule is defined by Equation 5.
 Now, in the signal constellation space S, set partitioning is designed by employing the rules described in Channel coding with multilevel/phase signals, IEEE Trans. Inform. Theory, vol. IT28, pp. 5567, 1982. One signal point in the space represents a sequence of channel output. Due to extensive channel memory, however, such a set partitioning process is more involved than memoryless channels with conventional multilevel/phase signaling. Realizing good distance properties between subsets is often realized by brute force computer searching. In particular, given the channel state S^{−} and input S of length L, the squared Euclidean distance between signal point y=(y_{i}, y_{i+1}, . . . ,y_{i+L−1}) and y′=(y′_{i}, y′_{i+1}, . . . , y′_{i+L−1}) is computed as
$\begin{array}{cc}d\left(y,{y}^{\prime}\right)=\sum _{i=0}^{L1}{\left({y}_{i}{y}_{i}^{\prime}\right)}^{2}.& \mathrm{Equation}\text{\hspace{1em}}7\end{array}$  During the partitioning process, the algorithm needs to calculate the distance between every pair of signal points in the associated subsets. Thus, the brute force computation complexity grows exponentially with the number of partition levels and the channel memory length.
 A systematic way of structured set partitioning (SSP) is described below using multiple embedded linear subspaces and their cosets.
 Let E^{L }be the set of all binary words of length L with components 0 and 1,
T(S ^{−})={Y=f(S ^{−} ⋄S), S∈E ^{L}} Equation 8  and T be the union of T(S^{−}) over S^{−}, i.e.,
$\begin{array}{cc}T=\bigcup _{S}T\left({S}^{}\right)=\left\{Y=f\left({S}^{}\diamond S\right),\left({S}^{}\diamond S\right)\in {E}^{m+L}\right\}& \mathrm{Equation}\text{\hspace{1em}}9\end{array}$  Thus, T(S^{−}) represents the set of signal points (in the L dimensional signal space S) originating from the channel state S^{−}, which is a subset of T(S) denoting the signal set of all channel output with length L. Since the mapping from E^{m+L }signal set T(S) is deterministic and onetoone, set partition can be performed on the linear space E^{m+L}.
 Formally, the partitioning is performed as follows:
 1. First level partition: over the whole linear space E^{m+L}, a subspace V of dimension k(1) is chosen, and L(1)=2^{k−k(1)}−1 cosets W_{1}, W_{2}, . . . , W_{L(1) }are constructed. In particular, the coset leaders of the cosets W_{1}, W_{2}, . . . , W_{L(1) }are a_{1}, . . . , a_{L(1)}, respectively. In other words,
W _{i} =V⊕a _{i} Equation 10
for i=1, . . . , L(1). The basic subspace V is the trivial coset W_{0 }and corresponds to the coset leader a_{0}={0, . . . , 0}∈V. Let$\begin{array}{cc}{T}_{i}=\left\{y=f\left(s\right),s\in {W}_{i}\right\}\text{}\mathrm{and}& \mathrm{Equation}\text{\hspace{1em}}11\\ {d}_{1}^{2}=\underset{i,j,i\ne j}{\mathrm{min}}\underset{y,z}{\mathrm{min}}\text{\hspace{1em}}d\left(y\in {T}_{i},z\in {T}_{j}\right)& \mathrm{Equation}\text{\hspace{1em}}12\end{array}$  where d(y,z) is the squared Euclidean distance between y and z as defined in Equation 7. To see that T_{i }is a partition of T, note that if d_{1} ^{2}>0, T_{i}∩T_{j}=∅ for all i≠j. Furthermore, it maintains that T=∪T_{i}.
 2. Second level partition: for each coset W_{i }of the first level partition, a new subspace V_{i }of dimension k(2)<k(1) is chosen in W_{i}. Subsequently, L(2)=2^{k(1)−k(2)}−1 cosets W_{i,1}, W_{i,2}, . . . , W_{i,L(2) }of V_{i }are constructed, which correspond to coset leaders a_{i1}, a_{i,2}, . . . , a_{i,L(2)}, respectively. V_{i }is the trivial coset with coset leader a_{i,0}=(0,0, . . . ,0)∈V. Similarly, by defining
$\begin{array}{cc}{T}_{i,j}=\left\{y=f\left(s\right),s\in {W}_{i,j}\right\},j=0,\dots \text{\hspace{1em}},L\left(2\right)\text{}\mathrm{and}& \mathrm{Equation}\text{\hspace{1em}}13\\ {d}_{2}^{2}=\underset{i}{\mathrm{min}}\underset{j,h,j\ne h}{\mathrm{min}}\underset{y,z}{\mathrm{min}}\text{\hspace{1em}}d\left(y\in {T}_{i,j},z\in {T}_{i,h}\right),& \mathrm{Equation}\text{\hspace{1em}}14\end{array}$  T_{i,j }is a partition of T_{i }and T.
 The partition can be carried on further to more levels following similar operations described above. For example, the subspace W_{i,j }at the second level can be further partitioned into W_{i,j,0}, W_{i,j,1}, . . . , W_{i,j,L(3)}, where L(3) is the number of subspaces that W_{i,j }contains and it holds that L(3)=2^{k(2)−k(3)}−1 and k(3) is the dimension of the subspaces on the third partition level.
 Once the set partition tree is completed, mapping from the input bits to transmitted channel bits (or SSP encoding) can be realized by assigning the branch address labels of the partition tree to the input bits accordingly. As a specific example, let us assume a three level twoway partition, i.e., L=3, L(1)=L(2)=L(3)=1, constructed on an ISI channel of length m+1=3. Given a channel state s^{−}=(s_{i−2}, s_{i−1}) and a block of input u=(u_{i}, u_{i+1}, u_{i+2}) of length 3, the input sequence can be encoded such that u_{i }represents the least significant bit which has a minimum squared Euclidean distance d_{1} ^{2 }while u_{i+2 }represents the most significant bit which has a minimum squared Euclidean distance d_{3} ^{2}. This is made possible by walking along the partitioning tree and assigning the branch label to the input bits. Specifically, it can be done in the following way. Given the input (u_{i}, u_{i+1}, u_{i+2}), first the coset W_{u} _{ 1 }, _{u} _{ i+1 }, _{u} _{ 1+2 }is chosen. Since there are exactly 8 cosets W_{i,j,k }at the third level, the assignment i=u_{i}, j=u_{i+1 }and k=u_{i+2 }can always be made. In each coset, W_{u} _{ i }, _{u} _{ i+1 }, _{u} _{ +2 }there are 4 binary words (s_{i−2}, s_{i−1}, s_{i}, s_{i+1}, s_{i+2}) corresponding to 4 different channel states s^{−}. The channel state s^{−}=(s_{i−2}, s_{i−1}) now uniquely determines the mapped outputs (s_{i}, s_{i+1}, s_{i+2}). This also defines the channel state for the next block of input bits.
 After the mapping, the (noise free) channel outputs corresponding to the block of input (u_{i}, u_{i+1}, u_{i+2}) possess the following properties:
 1. the minimum squared Euclidean distance in the 3dimensional output signal space is at least d_{1} ^{2 }for the signal outputs corresponding to input (u_{i}, u_{i+1}, u_{i+2}) and (u_{i}′, u_{i+1}′, u_{i+2}′, if u_{i}≠u_{i}′;
 2. the minimum squared Euclidean distance in the 3dimensional output signal space is at least d_{2} ^{2 }for the signal outputs corresponding to input (u_{i}, u_{i+1}, u_{i+2}) and (u′_{i}, u′_{i+1}, u′_{i+2}), if u_{i}=u′_{i}, but u_{i+1}≠u′_{i+1},
 3. the minimum squared Euclidean distance in the 3dimensional output signal space is at least d_{3} ^{2 }for the signal outputs corresponding to input (u_{i}, u_{i+1}, u_{i+2}) and (u′_{i}, u′_{i+1}, u′_{i+2}), if u_{i}=u′_{i}, u_{i+1}=u′_{i+1}, but u_{i+2}≠u′_{i+2}.
 4. furthermore, it holds that d_{1} ^{2}<d_{2} ^{2}<d_{3} _{2}.
 For practical applications, L(i)=1 for i=1, 2, . . . . In other words, subspaces are bisected into 2 smaller subspaces at each partition level. This on one hand simplifies the partitioning process; on the other hand, it facilitates the application of binary component codes when set partitioning is combined with multilevel coding, as will be explained later.
 To partition the linear space E^{m+L}, a simple parity check code can be used for successive partitioning. Each subspace W_{i }is partitioned by a parity check equation. The elements in W_{i }are classified into two subspaces depending on whether the element satisfies the particular parity check equation enforced at that partition level. The partition procedure is best illustrated via the following example.

FIG. 1B illustrates a first level set partition for PR channel with target response [1 11 12]. A 3level structured set partition is made for the partial response target [1 11 12].FIG. 1B shows the process of first level partition, where the parity check equation is given in the top box 152 and coset leaders 154, 156 are annotated in the respective branch. The PR channel has a memory of length m=2, which dictates the channel state is of length 2, i.e., S^{−}∈E^{2}. For a 3level partition, the input is grouped into blocks of length 3, i.e., S^{−}∈E^{3}. Hence, the partition is performed on the complete space E^{5}. The parity check equation h_{1}=[1 0 0 0 1] bisects the space E^{5 }into two subsets 158, 160, that are designated by W_{0 }and W_{1 }respectively. By using the notations from above, the basis subspace V (i.e., W_{0}) for the first level partition contains all the binary vectors S*ΔS^{−}⋄S which satisfy the parity check equation. That is,
W _{0} ={S*h _{1} ·S*=0}. Equation 15  As the coset leader for W_{1 }is given by a_{1}=[0 0 0 0 0 1], the other coset on the first level partition is
W _{1} ={S*h _{1} ·S*=h _{1} ·a _{1} ^{T}}. Equation 16  Calculation verifies that the squared Euclidean distance between the signal points corresponding to W_{0 }and W_{1 }is at least 4. That is,
$\begin{array}{cc}{d}_{1}^{2}=\underset{y,z}{\mathrm{min}}\text{\hspace{1em}}d\left(y\in {T}_{0},z\in {T}_{1}\right)\ge 4\text{}\mathrm{where}& \mathrm{Equation}\text{\hspace{1em}}17\\ {T}_{0}=\left\{y=f\left(s\right),s\in {W}_{0}\right\}\text{\hspace{1em}}\mathrm{and}\text{\hspace{1em}}{T}_{1}=\left\{z=f\left(s\right),s\in {W}_{1}\right\}.& \mathrm{Equation}\text{\hspace{1em}}18\end{array}$  The next level of partition involves dividing W_{0 }and W_{1 }further into subsets, starting with W_{0}. A parity check equation h_{2,0}=[0 1 0 1 0] is chosen to bisect W_{0 }into two subsets, depending on the parity check result is 0 or 1. Using the coset notion, vectors in the space W_{0,0 }satisfy
W _{0,0} {S*h _{2,0} ·S=0} Equation 19  in addition to Equation 15. For subset W_{0,1}, it becomes
W _{0,1} ={S*h _{2,0} ·S*=h _{2,0} a _{2} ^{T}} Equation 20  in addition to Equation 15. For ease of encoding/decoding, it is preferred to construct a set of parity check equations in a systematic format. To accomplish this, Gaussian elimination is adopted during the set partition construction if necessary. In this case, h_{1 }and h_{2,0 }are already in systematic form and no Gaussian elimination is needed.

FIG. 2 illustrates a second level set partition for W_{0 } 202 for PR channel [1 11 12].FIG. 2 illustrates the two equivalent forms 204, 206 of partition for W_{0}. The bottom half ofFIG. 2 (at 206) contains the two parity check equations and associated coset leaders to arrive at W_{0,0 }at 208 and W_{0,1 }at 210 from E^{5}. 
FIG. 3 illustrates a second level set partition for W_{1 }at 302 for PR channel [1 11 12]. The bottom partition 304 is an equivalent representation of the top partition 306 in a systematic form.  Similarly, W_{1 }can be partitioned into two subsets W_{1,0 }at 308 and W_{1,1 }at 310, as illustrated in
FIG. 3 . The only difference from the partition of W_{0 }is that the coset leaders a_{1 }corresponding to W_{1 }is [0 0 0 0 1], in contrast to allzero vector for W_{0}. Define
T _{i,j} ={y=f(s), s∈W _{ij}, i,j=}0,1 Equation 21
It holds that$\begin{array}{cc}{d}_{2}^{2}=\underset{i}{\mathrm{min}}\underset{j,h,j\ne h}{\mathrm{min}}\underset{y,z}{\mathrm{min}}\text{\hspace{1em}}d\left(y\in {T}_{i,j},z\in {T}_{i,h}\right)\ge 408.& \mathrm{Equation}\text{\hspace{1em}}22\end{array}$  The intersubset squared Euclidean distance grows from d_{1} ^{2}=4 on the first level to d_{2} ^{2}=408 on the second level, providing a good distance spectrum for multilevel encoding, as explained in more detail below.

FIG. 4 illustrates third level set partition for W_{0,0 }at 402 for PR channel [1 11 12]. The bottom partition 404 is an equivalent representation of the top partition 406 in systematic form.  It is a straightforward repetition of the above procedures to further partition the subsets W_{0,1}, W_{1,0}, and W_{1,1 }into smaller subsets. The following description applies to partition of W_{0,0}. The parity check equation and its equivalent form for partitioning W_{0,0 }is depicted in
FIG. 4 . Depending on whether s* satisfies the parity check equation
h _{3}·(s*)^{T}=0 Equation 23  the subspace W_{0,0 }is partitioned into W_{0,0,0 }and W_{0,0,1 }with
W _{0,0,1} =W _{0,0,0} ⊕a _{4} Equation 24
where h_{3}=[0 1 1 0 0] and a_{4}=[0 0 1 0 0]. Hence, the equivalent parity check matrix becomes$\begin{array}{cc}{H}_{3,0}\stackrel{\u25b3}{=}\uf603\begin{array}{c}{h}_{3}\\ {h}_{2}\\ {h}_{1}\end{array}\uf604=\uf603\begin{array}{ccccc}0& 1& 1& 0& 0\\ 0& 1& 0& 1& 0\\ 1& 0& 0& 0& 1\end{array}\uf604.& \mathrm{Equation}\text{\hspace{1em}}25\end{array}$  The associated coset leader for W_{0,0,1 }is a_{4}=[0 0 1 0 0].
 The partition of W_{0,1}, W_{1,0 }are the same as W_{0,0}. However, partition of W_{1,1 }is different. The corresponding parity check equation is instead
h′ _{3}·(s*)^{T}=0, Equation 26  and the associated coset leader remains a_{4}=[0 1 0 0].

FIG. 5 illustrates a three level structured set partition 500 for PR channel [1 11 12]. Let$\begin{array}{cc}{T}_{i,j,h}=\left\{y=f\left(s\right),s\in {W}_{i,j,h}\right\},\text{\hspace{1em}}i,j,h=0,1\text{}\mathrm{and}& \mathrm{Equation}\text{\hspace{1em}}27\\ {d}_{3}^{2}=\underset{i,j}{\mathrm{min}}\underset{g,h,g\ne h}{\mathrm{min}}\underset{y,z}{\mathrm{min}}\text{\hspace{1em}}d\left(y\in {T}_{i,j,g},z\in {T}_{i,j,h}\right).& \mathrm{Equation}\text{\hspace{1em}}28\end{array}$  it can be shown that with the complete 3level partition 500, it holds that
d _{3} ^{2}≧1064>d _{2} ^{2}≧328>d _{1} ^{2}=4. Equation 29  Therefore, {d_{1} ^{2}=4, d_{2} ^{2}=328, d_{1} ^{2}=1064} is the hierarchy of minimal squared Euclidean distances for the 3level set partition 500.
 Given a channel state s^{−}=s_{−1},s_{−2 }and a block of input u=(u_{i},u_{i+1},u_{i+2}) of length 3, the input sequence can be encoded such that u_{i }represents the least significant bit which has a minimum squared Euclidean distance d_{1} ^{2 }while u_{i+2 }represents the most significant bit which has a minimum squared Euclidean distance d_{3} ^{2}. This is possible by assigning the branch label of the SSP at the first level 502, the second level 504 and the thirdlevel 506 in
FIG. 5 to u_{0}, u_{1 }and u_{2}, respectively. Let$\begin{array}{cc}H\stackrel{\u25b3}{=}\uf603\begin{array}{ccccc}0& 1& 1& 0& 0\\ 0& 1& 0& 1& 0\\ 1& 0& 0& 0& 1\end{array}\uf604=\left[{H}_{s},I\right],& \mathrm{Equation}\text{\hspace{1em}}30\\ {H}_{s}=\uf603\begin{array}{cc}0& 1\\ 0& 1\\ 1& 0\end{array}\uf604,& \mathrm{Equation}\text{\hspace{1em}}31\\ {H}^{\prime}\stackrel{\u25b3}{=}\uf603\begin{array}{ccccc}0& 0& 1& 0& 0\\ 0& 1& 0& 1& 0\\ 1& 0& 0& 0& 1\end{array}\uf604=\left[{H}_{s}^{\prime},I\right],& \mathrm{Equation}\text{\hspace{1em}}32\\ {H}_{\text{\hspace{1em}}s}^{\text{\hspace{1em}}\prime}=\uf603\begin{array}{cc}0& 0\\ 0& 1\\ 1& 0\end{array}\uf604,& \mathrm{Equation}\text{\hspace{1em}}33\\ A\stackrel{\u25b3}{=}\uf603\begin{array}{c}\begin{array}{c}{a}_{1}\\ {a}_{2}\end{array}\\ {a}_{3}\end{array}\uf604=\uf603\begin{array}{ccccc}0& 0& 0& 0& 1\\ 0& 0& 0& 1& 0\\ 0& 0& 1& 0& 0\end{array}\uf604=\left[\begin{array}{cc}0& {A}_{s}\end{array}\right]& \mathrm{Equation}\text{\hspace{1em}}34\\ {A}_{s}=\uf603\begin{array}{ccc}0& 0& 1\\ 0& 1& 0\\ 1& 0& 0\end{array}\uf604.& \mathrm{Equation}\text{\hspace{1em}}35\end{array}$  Note that rows of A are coset leaders at the 1st, 2nd and 3rd levels of partition. From
FIG. 5 , the encoding from {u_{0}, u_{1}, u_{2}} to the channel bits {s_{0}, s_{1}, s_{2}} can be verified to take the form
[s _{0 } s _{1 } s _{2} ]=[s _{−2 } s _{−1} ]·H _{s} ^{T} ⊕[u _{0 } u _{1 } u _{2} ]·A _{s }ifu _{0}&u _{1} =1 Equation 36  and
[s _{0 } s _{1 } s _{2} ]=[s _{−2 } s _{−1} ]·H′ _{s} ^{T} ⊕[u _{0 } u _{1 } u _{2} ]·A _{s }ifu _{0}&u _{1} =1 Equation 37  The encoding of Equations 36, 37 can be simplified to
s _{0}=(s _{−1}&(u_{0}&u _{1}) )⊕u _{2} Equation 38
s _{1} =s _{−1} ⊕u _{i} Equation 39
s _{2} =s _{−2} ⊕u _{o} Equation 40  Correspondingly, the decoder can be implemented by the following logic
u _{0} =s _{−2} ⊕s _{2} Equation 41
u _{1} =s _{−1} ⊕s _{1} Equation 42
u _{2}=(s _{−1}&(s _{−2} ⊕s _{2})&(s _{−1} ⊕s _{1}) )⊕s _{0} Equation 43  For the given example, the encoding and decoding for coded modulation is very simple.
 In general, when the target length becomes long (>3) and partition level increases, the encoding and decoding logic may become more complex. Under such cases, the endec can be realized via table lookup. Table 1 shows one possible structure of the encoding and decoding lookup table. Encoding can be executed as follows:
 1. Find the current channel state S^{−} from the encoder output as a result of last block of input;
 2. Locate the entrance corresponding to the channel state S^{−} in the ENDEC table.
 3. Find the row in the ENDEC table whose “User Bits” column contains the current input data block;
 4. The encoding output is the entrance under “Channel Bits”.
 The decoding operation can be performed in a similar way. The only difference is that the “input” is “Channel Bits” and the output becomes “User Bits”.
TABLE 1 ENDEC lookup table User Bits Channel Bits Channel State S^{−} (u_{0}, u_{1}, u_{2}, . . .) (s_{0}, s_{1}, s_{2}, . . .) 00 . . . 000 . . . 000 . . . 001 . . . 101 . . . . . . . . . 01 . . . 000 . . . 010 . . . . . . . . . 10 . . . 000 . . . 100 . . . . . . . . . 11 . . . 000 . . . 110 . . . . . . . . . . . . . . . . . . . . .  After encoding, the (noise free) channel outputs for a block of input (u_{0},u_{1},u_{2}) possess the following properties:
 1. the minimum squared Euclidean distance d_{1} ^{2 }in the 3dimensional output signal space is at least 4 for the signal outputs corresponding to input (u_{0},u_{1},u_{2}) and (u′_{0},u′_{1}, u′_{2}), if u_{0},≠u′_{0};
 2. the minimum squared Euclidean distance d_{2} ^{2 }in the 3dimensional output signal space is at least 408 for the signal outputs corresponding to input (u_{0},u_{1},u_{2}) and (u′_{0},u′_{1}u′_{2}), if u_{0}=u′_{0}, but u_{1}≠u′_{1};
 3. the minimum squared Euclidean distance d_{3} ^{2 }in the 3dimensional output signal space is at least 1064 for the signal outputs corresponding to input (u_{0},u_{1},u_{2}) and (u′_{0},u′_{1},u′_{2}), if u_{0}=u′_{0}, u_{1}=u′_{1}, but u_{2}≠U′_{2}.
 Via proper set partitioning, user information bits can contain different levels of significance. For example, for 3level SSP, a block of user bits (u_{0},u_{1},u_{2}) have 3 different levels of error probability if there is no other outer coder imposed besides SSP. After SSP, the first bit u_{0 }has the least significance and it is more prone to error than u_{1 }and u_{2}. The last bit u_{2 }has the most significance and it is more immune to error than both u_{0 }and u_{1}. The gist of multilevel coding is to protect each bit u_{i }differently via individual binary code C^{i }according to their significance.

FIG. 6 illustrates a three level structured set partition 600 with general multilevel coding. Coded bits 602 are transmitted columnwise to the channel. In this coding structure, there are 3 different codes (denoted by C^{0}, C^{1 }and C^{2}, respectively), which are used to encode the bits in the first level (U_{0}), the second level (U_{1}) and the third level (U_{2}) respectively. The generated redundancy bits U_{c0}, U_{c1 }and U_{c2 }are concatenated with information bits U_{0}, U_{1 }and U_{2 }respectively. Subsequently, a block of 3 bits is formed by taking one bit from (U_{0}, U_{c0}), (U_{1}, U_{c1}) and (U_{2}, U_{c2}) sequentially, as indicated by the arrows 604, 606, 608 inFIG. 6 . The block of 3 bits are now encoded by the SSP logic, the result of which are transmitted to the recording circuitry. In retrospect, the bits in (U_{0}, U_{c0}), (U_{1}, U_{c1}) and (U_{2}, U_{c2}) experience monotonically increasing reliability during transmission due to SSP; however, the unequal error probability at different partition levels is balanced via different outer codes applied upon (U_{0}, U_{c0}), (U_{1}, U_{c1}) and (U_{2}, U_{c2}).  From the system level, one can consider the combined SSP and MLC process as a code C, which operates on block of information bits U_{0}, U_{1 }and U_{2}. The code rate R of the code C is given by
R=(r _{0} +r _{1} +r _{2})/3 Equation 44  where r_{0}, r_{1 }and r_{2 }are the code rate of code C^{0}, C^{1 }and C^{2 }respectively. In general, for an Llevel SSP and MLC, the overall code rate is given by
R=(l(0)k _{o} +l(1)k _{1} + . . . +l(L−1)k _{L−1})/N. Equation 45  Here, 2^{l(i)}, i=0, . . . , L−1 are the number of subspaces at the ith level originating from a single subspace at the (i−1)th level; k_{i}, i=0, . . . , L−1 are the number of q_{i}=2^{l(i)}ary information symbols of the code C^{i}; N is the total in a codeword. The minimum distance of code C observes the following lower bound. If the structured set partition has a minimum squared Euclidean distance spectrum {d_{0} ^{2}<d_{1} ^{2}< . . . <d_{L−1} ^{2}} then the minimum squared Euclidean distance d^{2}(C) of the Llevel concatenated code C satisfies the lower bound
d ^{2}(C)≧min{d _{0} ^{2} D _{0} ,d _{1} ^{2} D _{1}1, . . . ,d _{L−1} ^{2} D _{L−1}}, Equation 46  where D_{i }is the minimum Hamming distance of the ith level code C^{i}, i=0, 1, . . . , L−1.

FIG. 7 illustrates multilevel encoding and decoding with structured set partitioning. An encoder 702 and a decoder 704 for MLC with SSP are illustrated coupled to a binary saturated communication medium 706 such as a data storage disc in a disc drive. User data at 710 is first converted to L parallel branches 712, 714, . . . , 716 by a serial to parallel converter 718. Each branch of data 712, 714, . . . , 716 is encoded by the component encoder C^{i }at 720, 722, 724. The outputs of the component encoder C^{i } 720, 722, 724 are mapped by the setpartition logic 726 for signal modulation. Finally, the mapped data 728 are converted back to serial form X and transmitted to the medium 706. The decoding operation in decoder 704 takes in place in the reverse order of encoding.  The selection of component codes for multilevel encoding after SSP is important. One important aspect is the assignment of code rates to the individual coding levels. Based on various informationtheoretical parameters, such as capacity and coding exponent, of each equivalent channel at individual partition level, various design rules are established. For partial response channels, the most relevant design rule is the balanced distances rule. The rule states that the component codes can be chosen such that
d _{i} ^{2} ·D _{i}=constant, i=0, . . . ,L−1, Equation 47  where d_{i} ^{2 }is the squared Euclidean distance between cosets as defined before and D_{i }is the Hamming distance of the component code at the ith partition level. To limit the decoding and implementation complexity, this example is limited to simple parity check codes. Other more sophisticated codes, however, can also be used.
 Magnetic recording channels experience electronics noise and medium noise, among many other system disturbances. When channel normalized density is low, an appropriately chosen generalized PR (GPR) target can achieve near optimal performance with minimum noise enhancement due to channel equalization. GPR system can be alternatively implemented as noise predictive maximum likelihood (NPML). Comparing to direct GPR implementation, NPML facilitates circuit implementation of equalization and Viterbi detection (or its suboptimal variants like postprocessing). To account for medium noise, which is known to be data dependent, NPML detection can be extended to be data dependent, giving rise to the socalled patterndependent noise predictive (PDNP) detection. PDNP detectors utilize data dependent noise whitening filters to account for the datadependence of medium noise during Viterbi detection. While it demands more complexity than NPML detectors, PDNP detectors offer significant performance improvement over NPML when medium noise becomes dominant.
 For SSP, noise predictive detection provides another advantage. In particular, the realizable distance spectrum of coded modulation systems is closely related to specific channel response, i.e., nominal targets. By employing PDNP or NPML detection, one gains flexibility in choosing frontend equalization target that has good distance spectrum for structured set partition. However, it should be pointed out that the equalization target is not the effective target for Viterbi detection if noise prediction is involved. Indeed, it is not difficult to show that the effective targets translate into the convolution of the equalization target with the noise whitening filter(s). For PDNP detectors, effective targets become data dependent. To design the optimal SSP for data dependent targets, however, turns out to be computationally prohibitive and it provides little insight into the problem. Here, an adhoc method is used, where various “extended” targets are tested with SSP and subsequently BER and SFR results are obtained and compared in order to find a “good” SSP.
 An approach to find a good SSP for PDNP detectors is described below:
 1. At a desired operating point, find the noise whitening filters for a given frontend equalization target.
 2. Convolve the equalization target and noise whitening filters and obtain effective detection target(s).
 3. Choose an “experimental target” that is close to the effective detection targets for SSP design.
 4. Find a good SSP for the experimental target via an iterative process, where for a predefined distance spectrum an exhaustive search is performed by computer for SSP. The exhaustive search is possible due to the fact that for practical applications, dimensions are limited to a manageable range, e.g., 6. If for a predefined distance spectrum an SSP can be found, the distance spectrum is further increased until there is no SSP can be found. The last available partitioning and the associated spectrum are then employed for SSP encoding.
 Recording channels suffer signaltonoise ratio degradation approximately proportional to R^{2}, in contrast to R for other communication channels. Here, R is the overall coderate. Hence, highrate codes are favored in recording channel applications. If SSP/MLC are employed, there exist various choices for component codes at each individual partition level. The detailed discussion here is limited to simple parity check codes as component codes. In particular, single parity check codes are employed for the bits having the smallest squared Euclidean distances as a result of SSP. This group of bits are most prone to error due to the shortest distance between the corresponding cosets. Bits that correspond to lower partition levels are left uncoded. The reasons to adopt such a coding structure can be summarized as in the following:
 1. Single parity codes are suitable for maximumlikelihood (ML) sequence detection without incurring significant complexity increase for intersymbol interference channels. For block parity codes, each additional parity check equation implies doubling the trellis size for ML detection. Hence, single parity block codes are used for a practical implementation.
 2. Single parity codes can have very high code rate. For coderate sensitive recording channels, high coderate is often preferred. In addition, the error statistics of single parity check codes are more compatible to outer ReedSolomon codes from a sectorfailurerate perspective.
 To apply SSP/MLC to recording channels, there are many other issues that need to be properly taken care of. One important issue is the compatibility of SSP mapping with other channel codes such as runlength limited (RLL) codes, DCfree (DCF) codes and so forth. These constrained codes are necessary for read channel system to work properly. At a first glance, it seems that SSP conflicts with other channel codes as SSP requires bits mapping which alters the property of the input sequence. However, as explained below, SSP can be made to coexist with other channel codes via a technique dubbed “reverse coding”. Secondly, in order to achieve and exploit good distance spectrum for SSP, typically up to 4 levels of set partition is desired for ISI channels. When SSP combined with channel memory, the trellis size often increases considerably, making SSP less attractive for practical application. As shown below, however, under most cases SSP can take advantage of the existing expanded trellis owing to noise prediction without demanding further trellis augmentation.

FIG. 8 illustrates a SSP/MLC encoder 802 that results in minimum disturbances to other constrained code encoding such as ECC & RLL/DCF encoding at encoders 804. InFIG. 8 , an encoder 800 for encoding user data 801 comprises a DCfree encoder 809 that receives a block of user data 801 and provides a block of DCfree encoded data output 803.  The encoder 800 comprises a rate1 mapper 806 (also called SSP decoder 806) that receives the block of DCfree encoded output 803. The rate1 mapper 806 performs a reverse mapping based on structured set partitioning, and provides an intermediate data sequence 808.
 The encoder 800 comprises a multilevel encoder 810 that receives the intermediate data sequence 808 and that generates redundant bits 812 based on the intermediate data sequence 808.
 The encoder 800 comprises a multiplexer 816 that concatenates the redundant bits 812 with the block of DCfree encoded data output 803. The multiplexer 816 provides an encoder output 817 that is encoded for transmission through a binary medium 818.
 The DCfree encoder 804 preferably comprises a runlengthlimited encoder. The rate1 mapper 806 arranges the block of DCfree encoded data output into a first matrix form by rows and columns, and generates the intermediate data sequence arranged as a second matrix. The rate1 mapper 806 uses a structured set partition of signals of an intersymbol interference channel, and generates the second matrix with rows and columns that have differentiated reliabilities after transmission through in a noisy intersymbol interference channel. The structured set partitioning preferably comprises multiple level partitioning of a linear space, and bisection of spaces into two subspaces at each partitioning level, where each subspace is a coset of the other subspace at the same partitioning level. The multilevel encoder generates redundant bits conforming to the differentiated reliabilities of the rows and columns of the intermediate data sequence, producing an optimized minimum Euclidean distance between output sequences of an intersymbol interference channel. The concatenator 816 concatenates the redundant bits from multiple blocks of the DCfree encoded data outputs into a redundant bits sequence, and appends the redundant bits sequence to the multiple blocks of DCfree encoded data outputs and subsequently transmits output 817 to the medium 818.
 SSP mapping essentially is a rate1 encoder that maps the input bits into channel bits such that the input bits possess ordered error probability. The ordered (monotonically increasing intrablock) reliability is then balanced via levelled component codes. Viewed from a different angle, as long as the component codes operate on the bits that have ordered reliability, SSP with MLC is achieved. Such an interpretation provides a SSP/MLC encoding scheme as illustrated in
FIG. 8 .  During encoding, the user bits are first encoded by other constrainedcode encoder such as RLL, DCF encoders 804. The encoded bits are then “decoded” by a SSP decoder 806. Since the RLL/DCF encoded bit sequence are directly transmitted to the channel, the SSP decoded bits (“intermediate bits”) at 808 possess ordered reliability. In other words, these intermediate bits can be considered as the user bits {u_{i}} as before, which are SSP encoded and subsequently transmitted to the channel. Hence, the MLC component codes provided by MLC encoder 810 can operate on the intermediate bitsequence 808 to balance the error probability on individual partition levels. The redundancy bits (parity bits) 812 generated by the MLC encoder 810 are appended to the encoded bit sequence 814 by concatenater 816 for transmission to the medium 818. Via such an encoding scheme, minimum disturbance is resulted from the SSP/MLC “encoding” and the bit sequence can still satisfy the RLL/DCF constraint, which may be slightly relaxed due to the insertion of parity bits. In fact, via the encoder structure in
FIG. 8 , it is possible to integrate the RLL/DCF and SSP/MLC encoder in one entity without much difficulty. This technique is referred to here as “reverse encoding” where the encoding is actually realized via a dummy decoding operation of the channel bits. 
FIG. 9 illustrates a block diagram of a decoder for SSP/MLC and RLL/DCF. The corresponding encoder is illustrated inFIG. 8 . A received (equalized) noisy signal is first processed by a channel sequence detector 902 and a detected bit sequence 904 is then punctured away from any parity bits generated during MLC encoding by puncturer 906. A punctured bit sequence 908 is subsequently decoded by the RLL/DCF decoder 910, followed by ECC decoding at ECC decoder 912. If single parity codes are used for MLC, the channel sequence detector is embedded with an SSP decoder since the parity is calculated on the decoded intermediate bits.  The decoder 900 comprises a channel detector 902 that provides an encoded user data output 904. The decoder 900 comprises a puncturer 906 that receives the encoded user data output 904. The puncturer removes parity bits, and provides a punctured data output 908. The decoder 900 comprises a DCfree decoder 910 that receives the punctured data output 908. The DCfree decoder 910 removes DCfree encoding, and providing a decoded user data output. The decoder 900 comprises an errorcorrectioncode decoder 912 that receives the decoded user data output, removes errors, and provides an errorcorrected user data output.

FIG. 10 illustrates a sequence detection for channel bits that are SSP/single parity encoded. The parity check equations are operated on the SSP decoded sequence (intermediate data sequence). In the diagram 950, an example of overall coderate 63/64 single parity bit MLC code is shown. The SSP level is 4, and the single parity is operated only on the address bits of the first partition level. Once the sequence detector reaches the boundary of the parity check equation, the competing paths are examined to see if they belong to a codeword. This can be accomplished via first translating the path bits into decoded bits by an embedded SSP decoder, and then the parity equation is inspected. Competing paths that do not satisfy parity check equations are discarded at the check points 952, 954, 956 which coincide with the parity check boundaries. Other operations in diagram 950 are similar to conventional Viterbi algorithm, with the exception that noise prediction can be also embedded for branch metric calculation.  Consider an ISI channel of length m+1 staged with an SSP of level L. The corresponding signal space has a dimension m+L, where the set partition is operated. Consequently, the SSP encoding and decoding need to collect m+L user or channel bits at one time for appropriate operation. Such a requirement poses significant challenges for signal detection and ML decoding. Due to the presence of ISI as well as SSP memory, the transitions associated with the decoding trellis must collect m+L channel bits for decisionfeedback free maximumlikelihood detection. As a result, the number of states for the decoding trellis is at least 2^{m+L−1}. In addition to ISI and SSP, MLC encoding calls for additional trellis augmentation. In particular, with single parity check coding, the size of decoding trellis is twice that of the original size, i.e., with 2^{m+L }number of states. Under typical operating conditions, this amounts to 128 states for a the decoding trellis, making SSP/MLC less attractive for practical highspeed implementations. An empirical observation of SSP encoding/decoding which can be taken advantage of during implementation is described below.

FIG. 11 illustrates a process during SSP decoding. A first bit on which the parity check equation is effective is independent of the mth bit in the channel memory for most PR targets. The decoding trellis thus only requires 2^{m+L+1 }number of states for ML decoding.  Single parity check codes as component codes for MLC are preferred, although other more sophisticated codes can be used. In particular, the single parity check is operated on the bits corresponding to the address bits at the first partition level, since these bits are most susceptible to error. In most practical applications, it can be verified via computer that the SSP decoding results of the bits at the first partition level are independent of the mth bit in the memory. This relationship is explained in
FIG. 11 , which illustrates the SSP decoding operation with ISI memory m and partition level L. The SSP decoding is carried out by simply locating the coset corresponding to the channel bits and finding the corresponding address bits of the coset. The interesting property with SSP decoding is that, for most PR targets, the first bit after SSP decoding is independent of the mth bit in the channel memory, both of which are shaded inFIG. 11 . Since the parity check equation is only effective on the first bit, this implies that in fact one only needs to collect m+L−1 bits excluding the mth bit in the memory for ML decoding. Consequently, the decoding trellis only requires 2^{m+L−1 }number of states for an Llevel SSP partition of an ISI channel of length m+1, comparing to an otherwise 2^{m+L }number of states.  Presented below is an SSP/MLC architecture for perpendicular recording read channels. In order for the application of SSP/MLC, the following constraints are accomplished from the complexity and system performance perspective:
 1. ML decoding trellis size falls into a reasonable range, e.g., 32 states or 64 states;
 2. SSP/MLC is compatible with RLL and DCF encoding;
 3. Performance gain can be obtained at sector failure rate (SFR) level.
 Investigation of SSP/MLC for recording channels suggests that an L=4 level SSP is usually desired for a balanced tradeoff between performance and complexity. With a frontend equalization target of length m=3, such a configuration suggests a trellis size of 2^{m+L−1}=64 states, excluding any other augmentation effect resulted from MLC. However, by applying the technique presented below, it is possible to reduce the number of states to 32 states when the component codes operate only on the bits at the first partition level. Consequently, it becomes possible to limit the total number of states for the detection trellis to 64 states by employing a single parity check code as component codes.
 To account for the RLL/DCF constraint, the reverse coding technique described herein. With reverse coding, it is possible to concatenate SSP/MLC with RLL/DCF encoder in a straightforward manner, although at some cases the RLL/DCF constraint may be slightly altered due to the insertion of parity bits. With a more sophisticated design, it is possible to integrate SSP/MLC encoding with RLL/DCF. In the presentation below, the discussion is limited to concatenating SSP/MLC with an existing RLL/DCF encoder. The RLL/DCF encoder is considered as a “blackbox” and SSP/MLC encoding is achieved in a concatenated manner regardless of how the RLL/DCF encoding is achieved. This facilitates the channel design process as one can separate the design of RLL/DCF from SSP/MLC, and consequently many existing RLL/DCF encoders can be utilized.

FIG. 12 illustrates an SSP/MLC coded system 1202 with ECC encoding 1204 and RLL/DCF encoding 1206. Reverse coding technique is employed to preserve the RLL/DCF encoded sequence. The user bits are ECC encoded and subsequently subject to RLL/DCF encoding. The RLL/DCF encoded sequence is then again encoded by the SSP/MLC encoder 1202. As an example, the case of single parity as component codes after SSP is shown, where every 64 SSP decoded bits are appended with one single parity. As before, the single parity operates only on the first level of bits {s_{i}}. The computed parity bits (a) are then inserted back into the RLL/DCF encoded sequence and subsequently this newly formed sequence is sent to the channel for recording. In the receiver side, SSP/MLC decoding at 1208 is implemented via a modified Viterbi with embedded patterndependent noise prediction. The Viterbi detector is similar to the one shown inFIG. 10 . The detected bit sequence is then punctured where the SSP parity bit is removed and the reassembled sequence is then decoded by the RLL/DCF decoder 1210 and ECC decoder 1212.  For the simulation results presented below, a blackbox RLL/DCF encoder is used which has a code rate of 60/62. In order to preserve the RLL/DCF codeword boundary, the single SSP parity is computed on the 15 bits at the first partition level corresponding to the SSP decoded first 60 bits for each RLL/DCF codeword, and the left 2 bits enters directly into the parity check equation. The effective target for SSP design is fixed at [1 10 13 3] for various operating points. Although it is possible to optimize the effective target for each operating point for improved performance, at the SFR level it is observed that such optimization provides limited further performance improvement.
 To summarize, a channel coding architecture described above utilizes structured set partition (SSP) and multilevel coding (MLC), which is also called coded modulation. Structured set partition exploits the inherent structure of PR channel outputs and provides monotonically increasing Euclidean distance for the address bits labelling branches of the set partitioning tree. With the help of multilevel coding, the bits having smaller Euclidean distances are coded with larger Hamming distance component codes, hence component codes are potentially more efficiently employed. Consequently, the system minimum Euclidean distance and thus the performance can be improved. For magnetic recording channels, severe ISI and dominant medium noise consist the main system impairment source. Patterndependent noise predictive (PDNP) detection is an effective detection method for highdensity recording channels. Due to noise prediction, the effective targets become different from the equalization target and they are data dependent. In turn, set partitioning for PDNP detection becomes more involved. Effective targets are obtained that are convolutional results of the equalization target and noise whitening filters. Subsequently, the SSP is designed according to an approximated and truncated version of the effective targets. Although the design rule with such an approach seems rather ad hoc, it provides effective improvement on the system performance under many operating conditions. To realize maximumlikelihood (ML) detection, complexity considerations limit the component codes selection for MLC in conjunction with SSP. Parity codes are shown, although other more sophisticated codes can also be utilized. In particular, the cases are presented where single parity check equations are imposed only on the bits corresponding to labels at the first level of the set partitioning tree. Such a coding configuration confines the ML detection complexity while providing sound compatibility to outer RS code for recording channels. With SSP and single parity encoding, it is possible to obtain system performance gains over conventional single parity encoding where no SSP is involved at the sector failure level (SFR) for perpendicular recording channels. A major concern associated with coded modulation is the compatibility of SSP with other channel constrained encoding, such as runlength limited (RLL) and DCfree (DCF) encoding. A reverse coding technique is described to provide SSP/MLC encoding without violating other imposed constraints. The reverse coding technique attains the compatibility by creating dummy bits, which are the results of SSP decoding given an constrained input data sequence. The component codes operate on these dummy bits and the generated parities are transmitted along with the given input data sequence, hence producing minimum disturbances on the existing constraint. Indeed, it is even possible to integrate coded modulation encoding with RLL and DCF encoding. Such an integrated encoder can generate output bits satisfying a given desired constraint. It is possible to shrink the ML detection trellis size without incurring any system performance loss by observing the independence between the SSP encoded bits and some memory bits. In particular, under most cases of interest, the bits corresponding to the labelling bits on the first level of the set partition tree are independent of the “oldest” bit in the ISI memory. As a consequence, the trellis size can be reduced by half without incurring decision feedback for ML detection. Numerical simulations demonstrate that system performance gains can be obtained via MLC in conjunction with SSP, with minimum modifications to existing channel architectures. This is particularly true for perpendicular recording channels operating at high linear density experiencing dominated medium noise disturbances. Although the discussion is limited to simple parity check codes as component codes for coded modulation encoding, much more flexibility and potentially achievable gains can be expected by utilizing more powerful component codes in conjunction with SSP, including turbo codes, TPC, LDPC codes and the like and these codes can be treated as described herein.
 It is to be understood that even though numerous characteristics and advantages of various embodiments of the invention have been set forth in the foregoing description, together with details of the structure and function of various embodiments of the invention, this disclosure is illustrative only, and changes may be made in detail, especially in matters of structure and arrangement of parts within the principles of the present invention to the full extent indicated by the broad general meaning of the terms in which the appended claims are expressed. For example, the particular elements may vary depending on the particular application for the encoding and decoding system while maintaining substantially the same functionality without departing from the scope and spirit of the present invention. In addition, although the preferred embodiment described herein is directed to an encoding and decoding system for data storage devices, it will be appreciated by those skilled in the art that the teachings of the present invention can be applied to other binary communication channels, without departing from the scope and spirit of the present invention.
Claims (20)
Priority Applications (1)
Application Number  Priority Date  Filing Date  Title 

US11/263,128 US7205912B1 (en)  20051031  20051031  Structured set partitioning and multilevel coding for partial response channels 
Applications Claiming Priority (2)
Application Number  Priority Date  Filing Date  Title 

US11/263,128 US7205912B1 (en)  20051031  20051031  Structured set partitioning and multilevel coding for partial response channels 
CN 200610159851 CN1983432B (en)  20051031  20061030  Structured set partitioning and multilevel coding for partial response channels 
Publications (2)
Publication Number  Publication Date 

US7205912B1 US7205912B1 (en)  20070417 
US20070096950A1 true US20070096950A1 (en)  20070503 
Family
ID=37914140
Family Applications (1)
Application Number  Title  Priority Date  Filing Date 

US11/263,128 Active US7205912B1 (en)  20051031  20051031  Structured set partitioning and multilevel coding for partial response channels 
Country Status (2)
Country  Link 

US (1)  US7205912B1 (en) 
CN (1)  CN1983432B (en) 
Cited By (10)
Publication number  Priority date  Publication date  Assignee  Title 

US20110022921A1 (en) *  20090721  20110127  Ramot At Tel Aviv University Ltd.  Compact decoding of punctured block codes 
US20110022927A1 (en) *  20090721  20110127  Ramot At Tel Aviv University Ltd.  Compact decoding of punctured codes 
US20110022922A1 (en) *  20090721  20110127  Ramot At Tel Aviv University Ltd.  Compact decoding of punctured block codes 
US20110022920A1 (en) *  20090721  20110127  Ramot At Tel Aviv University Ltd.  Compact decoding of punctured block codes 
WO2013173708A1 (en) *  20120518  20131121  Micron Technology, Inc.  Methods and apparatuses for lowpower multilevel encoded signals 
US8625217B1 (en) *  20120927  20140107  Lsi Corporation  Branch metric computation and noise predictive calibration/adaptation for oversampled Y samples 
US20140173381A1 (en) *  20121219  20140619  Quantum, Inc.  Bit error detection and correction with error detection code and listnpmld 
EP2843662A1 (en) *  20130830  20150304  LSI Corporation  Systems and methods for multilevel encoding and decoding 
US9922686B2 (en)  20160519  20180320  Micron Technology, Inc.  Apparatuses and methods for performing intramodule databus inversion operations 
US10146719B2 (en)  20170324  20181204  Micron Technology, Inc.  Semiconductor layered device with data bus 
Families Citing this family (14)
Publication number  Priority date  Publication date  Assignee  Title 

US7406174B2 (en) *  20031021  20080729  Widevine Technologies, Inc.  System and method for ndimensional encryption 
JP2007215089A (en) *  20060213  20070823  Fujitsu Ltd  Decoding apparatus and decoding method 
US8091009B2 (en) *  20060323  20120103  Broadcom Corporation  Symbol by symbol map detection for signals corrupted by colored and/or signal dependent noise 
US7984359B2 (en) *  20060501  20110719  Seagate Technology, Llc  Correction of data errors in a memory buffer 
US7827464B2 (en) *  20061115  20101102  Seagate Technology Llc  Iterative read channel architectures with coded modulation 
JP5007676B2 (en)  20080131  20120822  富士通株式会社  Encoding apparatus, decoding apparatus, encoding and decoding apparatus and a recording and reproducing apparatus 
US7974037B2 (en) *  20080221  20110705  Hitachi Global Storage Technologies Netherlands, B.V.  Techniques for providing DCfree detection of DC equalization target 
KR101398200B1 (en) *  20080318  20140526  삼성전자주식회사  Memory device and encoding and/or decoding method 
JP5150325B2 (en) *  20080325  20130220  株式会社東芝  Xray detector 
US8677226B2 (en) *  20090504  20140318  Ikanos Communications, Inc.  Systems and methods for retransmission return channel error detection 
US8599959B2 (en) *  20101230  20131203  Lsi Corporation  Methods and apparatus for trellisbased modulation encoding 
GB2511493B (en) *  20130301  20170405  Gurulogic Microsystems Oy  Entropy modifier and method 
US10020912B2 (en)  20130313  20180710  Sans R&D, Llc  Method and a system for a receiver design in bandwidth constrained communication systems 
US10230392B2 (en) *  20161228  20190312  Intel Corporation  Techniques for parallel data decompression 
Citations (15)
Publication number  Priority date  Publication date  Assignee  Title 

US5412666A (en) *  19901109  19950502  Conner Peripherals, Inc.  Disk drive data path integrity control architecture 
US6014411A (en) *  19981029  20000111  The Aerospace Corporation  Repetitive turbo coding communication method 
US6704368B1 (en) *  19971128  20040309  Nokia Mobile Phones Limited  Coding and modulation method and apparatus for its implementation 
US6748079B1 (en) *  19981014  20040608  Macrovision Corporation  Method and system for recording data on and reading data from an optical medium 
US6751269B1 (en) *  19990111  20040615  Texas Instruments Incorporated  Bitinterleaved coded modulation for CATV upstream channels 
US20040117720A1 (en) *  20021122  20040617  Choi Eun A.  Pragmatic trellis code modulation decoder and a method thereof 
US20040136313A1 (en) *  20030115  20040715  Pctel, Inc.  Methods, apparatus, and systems employing soft decision decoding 
US20040139378A1 (en) *  20021030  20040715  Akhter Mohammad Shahanshah  Method and apparatus for error control coding in communication systems using an outer interleaver 
US20040205423A1 (en) *  20030313  20041014  Samsung Electronics Co., Ltd.  Data modulation method and apparatus capable of suppressing DC component using parity information of synchronization codeword 
US6829742B1 (en) *  19991027  20041207  Infineon Technologies Ag  Coding method and coding apparatus for coding a serial data stream 
US6865236B1 (en) *  20000601  20050308  Nokia Corporation  Apparatus, and associated method, for coding and decoding multidimensional biorthogonal codes 
US20050066261A1 (en) *  20030919  20050324  Fujitsu Limited  Method, apparatus, and circuit for record reproduction 
US20050149843A1 (en) *  20020531  20050707  Broadcom Corporation, A California Corporation  Bandwidth efficient coded modulation scheme based on MLC (multilevel code) signals having multiple maps 
US6917313B1 (en) *  20020116  20050712  Marvell International Ltd.  DCfree codes 
US6961010B2 (en) *  20030813  20051101  Seagate Technology Llc  DCfree code design with increased distance between code words 
Family Cites Families (1)
Publication number  Priority date  Publication date  Assignee  Title 

WO2003007300A1 (en)  20010709  20030123  Seagate Technology Llc  Method and apparatus for suppressing low frequency content in digital data 

2005
 20051031 US US11/263,128 patent/US7205912B1/en active Active

2006
 20061030 CN CN 200610159851 patent/CN1983432B/en active IP Right Grant
Patent Citations (15)
Publication number  Priority date  Publication date  Assignee  Title 

US5412666A (en) *  19901109  19950502  Conner Peripherals, Inc.  Disk drive data path integrity control architecture 
US6704368B1 (en) *  19971128  20040309  Nokia Mobile Phones Limited  Coding and modulation method and apparatus for its implementation 
US6748079B1 (en) *  19981014  20040608  Macrovision Corporation  Method and system for recording data on and reading data from an optical medium 
US6014411A (en) *  19981029  20000111  The Aerospace Corporation  Repetitive turbo coding communication method 
US6751269B1 (en) *  19990111  20040615  Texas Instruments Incorporated  Bitinterleaved coded modulation for CATV upstream channels 
US6829742B1 (en) *  19991027  20041207  Infineon Technologies Ag  Coding method and coding apparatus for coding a serial data stream 
US6865236B1 (en) *  20000601  20050308  Nokia Corporation  Apparatus, and associated method, for coding and decoding multidimensional biorthogonal codes 
US6917313B1 (en) *  20020116  20050712  Marvell International Ltd.  DCfree codes 
US20050149843A1 (en) *  20020531  20050707  Broadcom Corporation, A California Corporation  Bandwidth efficient coded modulation scheme based on MLC (multilevel code) signals having multiple maps 
US20040139378A1 (en) *  20021030  20040715  Akhter Mohammad Shahanshah  Method and apparatus for error control coding in communication systems using an outer interleaver 
US20040117720A1 (en) *  20021122  20040617  Choi Eun A.  Pragmatic trellis code modulation decoder and a method thereof 
US20040136313A1 (en) *  20030115  20040715  Pctel, Inc.  Methods, apparatus, and systems employing soft decision decoding 
US20040205423A1 (en) *  20030313  20041014  Samsung Electronics Co., Ltd.  Data modulation method and apparatus capable of suppressing DC component using parity information of synchronization codeword 
US6961010B2 (en) *  20030813  20051101  Seagate Technology Llc  DCfree code design with increased distance between code words 
US20050066261A1 (en) *  20030919  20050324  Fujitsu Limited  Method, apparatus, and circuit for record reproduction 
Cited By (18)
Publication number  Priority date  Publication date  Assignee  Title 

US8516352B2 (en)  20090721  20130820  Ramot At Tel Aviv University Ltd.  Compact decoding of punctured block codes 
US20110022927A1 (en) *  20090721  20110127  Ramot At Tel Aviv University Ltd.  Compact decoding of punctured codes 
US20110022922A1 (en) *  20090721  20110127  Ramot At Tel Aviv University Ltd.  Compact decoding of punctured block codes 
US20110022920A1 (en) *  20090721  20110127  Ramot At Tel Aviv University Ltd.  Compact decoding of punctured block codes 
US8375278B2 (en)  20090721  20130212  Ramot At Tel Aviv University Ltd.  Compact decoding of punctured block codes 
US8516351B2 (en)  20090721  20130820  Ramot At Tel Aviv University Ltd.  Compact decoding of punctured block codes 
US20110022921A1 (en) *  20090721  20110127  Ramot At Tel Aviv University Ltd.  Compact decoding of punctured block codes 
US9397699B2 (en)  20090721  20160719  Ramot At Tel Aviv University Ltd.  Compact decoding of punctured codes 
US9148170B2 (en)  20120518  20150929  Micron Technology, Inc.  Methods and apparatuses for lowpower multilevel encoded signals 
WO2013173708A1 (en) *  20120518  20131121  Micron Technology, Inc.  Methods and apparatuses for lowpower multilevel encoded signals 
US8854236B2 (en)  20120518  20141007  Micron Technology, Inc.  Methods and apparatuses for lowpower multilevel encoded signals 
US8625217B1 (en) *  20120927  20140107  Lsi Corporation  Branch metric computation and noise predictive calibration/adaptation for oversampled Y samples 
US20140173381A1 (en) *  20121219  20140619  Quantum, Inc.  Bit error detection and correction with error detection code and listnpmld 
US9047882B2 (en)  20130830  20150602  Lsi Corporation  Systems and methods for multilevel encoding and decoding 
EP2843662A1 (en) *  20130830  20150304  LSI Corporation  Systems and methods for multilevel encoding and decoding 
US9922686B2 (en)  20160519  20180320  Micron Technology, Inc.  Apparatuses and methods for performing intramodule databus inversion operations 
US10297294B2 (en)  20160519  20190521  Micron Technology, Inc.  Apparatuses and methods for performing intramodule databus inversion operations 
US10146719B2 (en)  20170324  20181204  Micron Technology, Inc.  Semiconductor layered device with data bus 
Also Published As
Publication number  Publication date 

CN1983432A (en)  20070620 
US7205912B1 (en)  20070417 
CN1983432B (en)  20100623 
Similar Documents
Publication  Publication Date  Title 

Immink et al.  Codes for digital recorders  
US8225148B2 (en)  Systems and methods for achieving higher coding rate using parity interleaving  
US7587657B2 (en)  Method and apparatus for iterative errorerasure decoding  
CN1279699C (en)  Method and coding apparatus using low density parity check codes for data storage or data transmission  
US8788916B2 (en)  Architecture and control of reedsolomon errorcorrection decoding  
US7516389B2 (en)  Concatenated iterative and algebraic coding  
US7725800B2 (en)  Decoding techniques for correcting errors using soft information  
US20040181732A1 (en)  Two stage detector having viterbi detector matched to a channel and post processor matched to a channel code  
US6005727A (en)  Servo decoder for decoding an error correcting servo code recorded on a disc storage medium  
US6516443B1 (en)  Error detection convolution code and post processor for correcting dominant error events of a trellis sequence detector in a sampled amplitude read channel for disk storage systems  
JP4833173B2 (en)  Decoder, encoding and decoding apparatus and a recording and reproducing apparatus  
US6216249B1 (en)  Simplified branch metric for reducing the cost of a trellis sequence detector in a sampled amplitude read channel  
US8151162B2 (en)  Encoding device, decoding device, encoding/decoding device, and recording/reproducing device  
US7743314B2 (en)  Method and apparatus for a datadependent noise predictive viterbi  
US8051363B1 (en)  Absorb decode algorithm for 10GBaseT LDPC decoder  
US7222289B2 (en)  Channel processor using reduced complexity LDPC decoder  
US6604220B1 (en)  Disk drive comprising a multipleinput sequence detector selectively biased by bits of a decoded ECC codedword  
US7805664B1 (en)  Likelihood metric generation for trellisbased detection and/or decoding  
US6597742B1 (en)  Implementing reducedstate viterbi detectors  
US20050273688A1 (en)  Data communication system with multidimensional errorcorrection product codes  
EP1267345A1 (en)  Information recording and reproducing apparatus and method and signal decoding circuit  
US8977941B2 (en)  Iterative decoder systems and methods  
US6804805B2 (en)  Method and apparatus for encoding with unequal protection in magnetic recording channels having concatenated error correction codes  
US6185173B1 (en)  Sampled amplitude read channel employing a trellis sequence detector matched to a channel code constraint and a post processor for correcting errors in the detected binary sequence using the signal samples and an error syndrome  
US6115198A (en)  PR4 sampled amplitude read channel for detecting user data and embedded servo data 
Legal Events
Date  Code  Title  Description 

AS  Assignment 
Owner name: SEAGATE TECHNOLOGY LLC, CALIFORNIA Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:YANG, XUESHI;KUZNETSOV, ALEXANDER;REEL/FRAME:017174/0682;SIGNING DATES FROM 20051024 TO 20051028 

STCF  Information on status: patent grant 
Free format text: PATENTED CASE 

AS  Assignment 
Owner name: JPMORGAN CHASE BANK, N.A., AS ADMINISTRATIVE AGENT Free format text: SECURITY AGREEMENT;ASSIGNORS:MAXTOR CORPORATION;SEAGATE TECHNOLOGY LLC;SEAGATE TECHNOLOGY INTERNATIONAL;REEL/FRAME:022757/0017 Effective date: 20090507 Owner name: WELLS FARGO BANK, NATIONAL ASSOCIATION, AS COLLATE Free format text: SECURITY AGREEMENT;ASSIGNORS:MAXTOR CORPORATION;SEAGATE TECHNOLOGY LLC;SEAGATE TECHNOLOGY INTERNATIONAL;REEL/FRAME:022757/0017 Effective date: 20090507 

FPAY  Fee payment 
Year of fee payment: 4 

AS  Assignment 
Owner name: SEAGATE TECHNOLOGY HDD HOLDINGS, CALIFORNIA Free format text: RELEASE;ASSIGNOR:JPMORGAN CHASE BANK, N.A., AS ADMINISTRATIVE AGENT;REEL/FRAME:025662/0001 Effective date: 20110114 Owner name: MAXTOR CORPORATION, CALIFORNIA Free format text: RELEASE;ASSIGNOR:JPMORGAN CHASE BANK, N.A., AS ADMINISTRATIVE AGENT;REEL/FRAME:025662/0001 Effective date: 20110114 Owner name: SEAGATE TECHNOLOGY LLC, CALIFORNIA Free format text: RELEASE;ASSIGNOR:JPMORGAN CHASE BANK, N.A., AS ADMINISTRATIVE AGENT;REEL/FRAME:025662/0001 Effective date: 20110114 Owner name: SEAGATE TECHNOLOGY INTERNATIONAL, CALIFORNIA Free format text: RELEASE;ASSIGNOR:JPMORGAN CHASE BANK, N.A., AS ADMINISTRATIVE AGENT;REEL/FRAME:025662/0001 Effective date: 20110114 

AS  Assignment 
Owner name: THE BANK OF NOVA SCOTIA, AS ADMINISTRATIVE AGENT, Free format text: SECURITY AGREEMENT;ASSIGNOR:SEAGATE TECHNOLOGY LLC;REEL/FRAME:026010/0350 Effective date: 20110118 

AS  Assignment 
Owner name: SEAGATE TECHNOLOGY US HOLDINGS, INC., CALIFORNIA Free format text: TERMINATION AND RELEASE OF SECURITY INTEREST IN PATENT RIGHTS;ASSIGNOR:WELLS FARGO BANK, NATIONAL ASSOCIATION, AS COLLATERAL AGENT AND SECOND PRIORITY REPRESENTATIVE;REEL/FRAME:030833/0001 Effective date: 20130312 Owner name: SEAGATE TECHNOLOGY LLC, CALIFORNIA Free format text: TERMINATION AND RELEASE OF SECURITY INTEREST IN PATENT RIGHTS;ASSIGNOR:WELLS FARGO BANK, NATIONAL ASSOCIATION, AS COLLATERAL AGENT AND SECOND PRIORITY REPRESENTATIVE;REEL/FRAME:030833/0001 Effective date: 20130312 Owner name: SEAGATE TECHNOLOGY INTERNATIONAL, CAYMAN ISLANDS Free format text: TERMINATION AND RELEASE OF SECURITY INTEREST IN PATENT RIGHTS;ASSIGNOR:WELLS FARGO BANK, NATIONAL ASSOCIATION, AS COLLATERAL AGENT AND SECOND PRIORITY REPRESENTATIVE;REEL/FRAME:030833/0001 Effective date: 20130312 Owner name: EVAULT INC. (F/K/A I365 INC.), CALIFORNIA Free format text: TERMINATION AND RELEASE OF SECURITY INTEREST IN PATENT RIGHTS;ASSIGNOR:WELLS FARGO BANK, NATIONAL ASSOCIATION, AS COLLATERAL AGENT AND SECOND PRIORITY REPRESENTATIVE;REEL/FRAME:030833/0001 Effective date: 20130312 

FPAY  Fee payment 
Year of fee payment: 8 

MAFP  Maintenance fee payment 
Free format text: PAYMENT OF MAINTENANCE FEE, 12TH YEAR, LARGE ENTITY (ORIGINAL EVENT CODE: M1553); ENTITY STATUS OF PATENT OWNER: LARGE ENTITY Year of fee payment: 12 