Patent Grant
7030789
The present invention relates to techniques for applying modulation constraints to data to eliminate data patterns prone to read errors, and more particularly, to techniques for ensuring that specific prohibited symbols do not occur periodically.
A disk drive can write data bits onto a data storage disk such as a magnetic hard disk. The disk drive can also read data bits that have been stored on a data disk. Certain data patterns are difficult to write onto a disk and often cause errors when the data patterns are read back. Long sequences of consecutive zeros or consecutive ones (e.g., 40 consecutive zeros) are examples of data patterns that are prone to errors. A long sequence of alternating polarity bits (010101010 . . . ) is another example of an error prone data pattern.
Therefore, it is desirable to eliminate error prone patterns in channel input data. Eliminating error prone patterns ensures reliable operation of the detector and timing loops in a disk drive system. One way to eliminate error prone data patterns is to substitute these data patterns with data patterns that are less likely to cause errors. The substitute symbols can be stored in memory in lookup tables. Lookup tables, however, are undesirable for performing substitutions of data patterns with a large number of bits, because they require a large amount of memory.
Many disk drives have a modulation encoder that uses modulation codes to eliminate error prone data patterns. Modulation encoders impose global and/or interleaved constraints on data to eliminate certain data patterns. A global G constraint at the input of a 1/(1+D2) precoder prohibits data patterns with more than G consecutive zeros. An interleaved I constraint at the input of a 1/(1+D2) precoder prohibits data patterns with more than I consecutive zeros in an even or odd interleave.
Modulation codes, also known as constrained codes, have been widely used in magnetic and optical data storage to eliminate sequences that are undesired for the processes of recording and reproducing digital data. Various classes of modulation codes are used in practice. For example, peak detection systems employing run length-limited RLL(d,k) constrained codes, such as rate-1/2 RLL(2,7) and rate-2/3 RLL(1,7) codes, have been predominant in digital magnetic storage at low normalized linear densities.
At moderate normalized linear densities, the introduction of partial-response maximum-likelihood (PRML) detection channels into data storage required a different type of constrained codes. This class of codes, which are known as PRML(G,I) codes, facilitates timing recovery and gain control, and limits the path memory length of the sequence detector, and therefore the decoding delay, without significantly degrading detector performance. PRML(G,I) codes are used in conjunction with 1/(1+D2) precoders and noise-predictive maximum likelihood (NPML) channels, which generalize the PRML concept.
More recently, maximum transition run (MTR) codes, in particular MTR(j,k) and MTR(j,k,t) codes, have been introduced to provide coding gain for noise-predictive maximum likelihood channels. MTR codes, which are used in conjunction with 1/(1+D) precoders, improve detector performance by eliminating or reducing the occurrence of dominant error events at the output of the sequence detector at the expense of increasing the baud rate and incorporating the j-constraint into the detector by implementing a time-varying trellis.
Many prior art techniques for using modulation codes to reduce the occurrence of errors require a high latency time. Therefore, it would be desirable to provide alternative modulation encoding techniques for eliminating error prone data patterns that reduce the latency time.
The present invention provides techniques for applying modulation constraints to data by replacing certain prohibited error prone data patterns using periodically changing symbol mappings. The techniques of the present invention are application to reverse concatenation encoding schemes.
Initially, user data in a first base is mapped to integers of a second base using a base conversion technique. The integers in the second base correspond to symbols.
Subsequently, periodically changing symbol mappings are performed during which prohibited symbols generated during base conversion are mapped to permitted symbols. The periodically changing symbol mappings occur in multiple phases, and the prohibited symbols are different in each phase. The resulting data is processed by a precoder in some embodiments.
According to one embodiment of the present invention, the input data is separated into even and odd interleaves. Radix conversion and then symbol substitution are performed on subsets of the even and odd interleaves. The resulting data is then recombined with the separated bits. The combined data is then processed by a precoder.
According to another embodiment of the present invention, a partial interleaver interleaves the Reed-Solomon (RS) parity symbols in between groups of the RS data symbols in a way that achieves the best modulation constraints of the resulting data.
Other objects, features, and advantages of the present invention will become apparent upon consideration of the following detailed description and the accompanying drawings, in which like reference designations represent like features throughout the figures.
Reverse concatenation is a coding technique that employs an efficient modulation encoder followed by a systematic Reed-Solomon (RS) encoder and a post-RS modulation encoder for RS parity symbols. Reverse concatenation architectures that have efficient pre-RS modulation codes may employ large block sizes. Reverse concatenation architectures do not suffer from error propagation, because the modulation decoder for these efficient codes follows the RS decoder. Soft information from the detector or inner parity decoder is readily provided to the ECC decoder.
The output stream of RS encoder 102 is encoded by a second modulation encoder block 103. The second modulation encoder can be, for example, an MTR(j,k,t) encoder with good error propagation properties followed by a 1/(1+D) precoder. Multiplexer 104 appends the output stream of block 103 to the modulation-encoded user data output of block 101. Parity insertion block 105 generates m inner parity bits as a function of n modulation-encoded bits and inserts them into the modulation-encoded data stream. The inner parity code rate of the output stream of block 105 is n/(n+m).
Partial interleaver block 114 partially interleaves RS parity symbols into the constrained encoded symbol stream from block 111. Partial interleaving is described in further detail below with respect to
Examples of modulation encoding schemes of the present invention that can be performed by modulation encoders 101, 103, and 111 are now discussed with respect to
At step 201, user data in the form of (n−1)-bit numbers [ai] at the input of the modulation encoder are converted from a first base into integers [bi] in a second base L using a base conversion technique. At step 202, periodically changing symbol mappings are performed to map the resulting symbols in the second base to constrained s-bit symbols [ci] using Boolean logic or lookup tables. The condition for the existence of a (n−1)/n rate code that belongs to this class is 2n−1≦Ln/s, where ‘s’ is the number of bits representing in each symbol of [bi]2.
The modulation encoder imposes a global constraint of G=26 and an interleaved constraint of I=22. In the absence of a precoder, these constraints are equivalent to limiting the maximum length of DC-patterns 00 . . . 0 and 11 . . . 1, and Nyquist patterns 0101 . . . and 1010 . . . to G+2 and the maximum number of consecutive like symbols in an even or odd interleave to I+1. These constraints can be abbreviated as PRML(G=26, I=22).
It should be understood the specific numbers used with respect to the particular parameters such as global G and interleaved I constraints, the base L, and the number of bits s in each symbol that are described with respect to
Radix conversion is performed at step 211. An example of a radix conversion technique that can be used in radix conversion step 211 is described in U.S. Pat. No. 6,236,340, which is incorporated by reference herein. Radix conversion is merely one example of a base conversion technique that can be performed according to the present invention. The discussion of radix conversion herein is illustrative and is not intended to limit the scope of the present invention.
During the radix conversion step 211, 199-bit words of user data are mapped into integers in base 991. The 199-bit binary integers are represented in
After radix conversion step 211, two periodically varying symbol mapping steps 212 and 213 are performed. Symbol mapping step 212 maps the base 991 integers [bi]2 generated by block 211 using a function f( ). An integer [bi]2 generated at step 211 has 20 symbols. Each symbol can be represented by 10 binary bits. Symbol mapping step 212 analyzes each 10-bit symbol in a [bi]2 integer.
Step 212 excludes 33 symbols that equal 0x0x0x0x0x or 1111111111 from the list of all possible symbols for [bi] in base 991, where x equals any binary value (1 or 0). Radix conversion step 211 can generate 991 possible symbols for each 991-base bit of a [bi]991 integer. A 10-bit binary number has 1024 possible values. Therefore, there are 991 possible symbols (1024−33=991) that remain for each of the 10-bit binary represented symbols of a [bi]2 integer.
If any of the 10-bit binary represented symbols of a [bi]2 integer match one of the prohibited symbols, it is replaced at step 212 with another 10-bit symbol that is not one of the 33 excluded symbols and that is not one of the 991 possible symbols that can be generated at radix step 211. Such a mapping is possible, because each 10-bit binary represented symbol has 1024 possible symbols, minus the 33 excluded symbols, which equals 991 possible symbols. The integers generated at step 212 are labeled [c0]2 through [c19]2 (the subscript 2 indicating a binary representation).
Thus, the range of the symbol mapping for the function performed at step 212 is a size-991 list of 10-bit symbols that is obtained by excluding 33 symbols of type 0x0x0x0x0x and 1111111111 from the list of all 10-bit symbols. The particular mapping of this function is preferably selected to minimize the complexity of the Boolean circuit implementing the function. Table 1 below lists the 33 symbols that are excluded in all of the four phases 0–3. Table 1, column 2 lists the symbols that are excluded at symbol mapping step 212 during phase 0 (i.e., i mod 4=0).
x0x0x0x0x0
x1x1x1x1x1
Step 213 performs periodically changing symbol substitutions on the integers [ci] generated at step 212. Step 213 performs symbol substitutions that vary in time during phases 1–3, depending on which symbol of [ci]2 is being considered.
Phase 0 occurs when the index i of a symbol of [ci]2 is evenly divisible by 4 (i.e., i mod 4=0). During phase 0, the excluded symbols are replaced at step 212, as described above. No action is taken in phase 0 during step 213 when the index of a [ci] symbol is 0, 4, 8, 12, 16, 24, etc. The symbols generated at step 213 are labeled [d0]2 through [d19]2.
Symbol substitutions are performed at step 213 for the [ci]2 symbols during phases 1, 2, and 3. The substitutions performed at step 213 during phases 1–3 are summarized in Table 2 below. Note that x in Table 2 stands for (x1, x2, x3, x4, x5), and ˜ stands for the Boolean negation. All the substitutions in phases 1–3 are relative to the list of allowed symbols in phase 0. Therefore, no substitution is performed in phase 0. No preceding operation is required for this particular embodiment. The periodically-varying symbol substitutions performed by step 213 are characterized by three mappings, g1(.), g2(.) and g3(.).
x
10x20x30x40x50
x
11x21x31x41x51
x ≠
x ≠
Phase 1 occurs each time the index i of a [ci] symbol has a remainder of 1 when divided by 4 (i.e., i mod 4=1). For example, phase 1 occurs when the indexes of a [ci] symbol are 1, 5, 9, 13, 17, 25, etc. During phase 1, the symbols shown in the second row, second column of Table 2 are replaced with the symbols shown in the third row, second column of Table 2. The substitute symbols are in the set of symbols (0x0x0x0x0x) excluded during phase 0 in step 212.
Phase 2 occurs each time the index i of a [ci] symbol has a remainder of 2 when divided by 4 (i.e., i mod 4=2). For example, phase 2 occurs when the indexes of a [ci] symbol are 2, 6, 10, 14, 18, 26, etc. During phase 2, the symbols shown in the second row, third column of Table 2 are replaced with the symbols shown in the third row, third column of Table 2. The substitute symbols are in the set of symbols (0x0x0x0x0x) excluded during phase 0 in step 212.
Phase 3 occurs each time the index i of a 10-bit block of a [ci] symbol has a remainder of 3 when divided by 4 (i.e., i mod 4=3). For example, phase 3 occurs when the indexes of a [ci] symbol are 3, 7, 11, 15, 18, 27, etc. During phase 3, the symbols shown in the second row, fourth column of Table 2 are replaced with the symbols shown in the third row, fourth column of Table 2. The substitute symbols are in the set of symbols (0x0x0x0x0x) excluded during phase 0 in step 212.
The modulation encoding technique of
The codes in this family have a symbol alphabet of size L=2s/2(2s/2−1)−1 and satisfy the constraints G=3s−4 and I=5(s/2)−3, where s is assumed to be even, and the code word size, n bits, is divisible by 4s. Although a precoder is not used for this class of codes, the constraints are still characterized by equivalent global G and interleaved I constraint values. In other words, the maximum length of DC- and Nyquist patterns at the recording channel input is G+2, and the maximum length of DC-patterns in an odd or even interleave of the recording channel input is I+1. Encoder 200 can also be used to construct an encoder that has a rate of 209/210 PRML(26,22) code with parameters n=210, L=991 and s=10 code, although 210 is not divisible by 40. In this case, the first symbol of a code word j occurs in phase (i mod 4), whereas in
At step 301, a base conversion technique is performed to convert user data in a first base into integers into a second base L representation [bi]. At step 302, periodically changing symbol mappings are performed to map the resulting base L symbols into constrained s-bit symbols [ci] using Boolean logic or lookup tables.
Precoder 303 then encodes the resulting constrained symbols after step 302. The precoding step removes prohibited symbols x1x1x1x1x1 and 1x1x1x1x1x. As a result, only two phases of symbol mappings are performed at step 302 to remove prohibited symbols 0x0x0x0x0x and x0x0x0x0x0.
Radix conversion step 311 of modulation encoder 310 uses a radix base-992 conversion of a 199-bit word at the encoder input. Step 312 performs symbol mappings of the [bi] symbols generated at step 311. Step 313 performs periodically varying symbol substitutions on the [ci] symbols generated at step 312. Step 314 performs 1/(1+D2) precoding on the [di] symbols generated at step 313.
Table 3 below lists the 32 symbols that are excluded in each of the two phases of encoder 310. The 2 phases are represented by the equation i mod 2. Symbol mapping step 312 performs a range of symbol mapping f(.) that is a size-992 list of 10-bit symbols obtained by excluding 32 symbols of type 0x0x0x0x0x from the list of all 1024 possible 10-bit symbols. If any of the 10-bit symbols of [bi]2 equals 0x0x0x0x0x, these symbols are mapped to another 10-bit symbol that does not equal 0x0x0x0x0x.
A 10-bit binary number has 1024 possible symbols. If 32 excluded symbols are subtracted from the 1024 possible symbols, 992 possible symbols remain. Each [bi]992 has 992 possible symbols. Therefore, there are enough symbols remaining to map each symbol that violates the constraint 0x0x0x0x0x to a symbol that does not violate this constraint and that is not generated in the radix conversion step. The particular mapping f(.) is preferably selected to minimize the complexity of the Boolean circuit implementing f(.). As can be seen from Table 3, symbol mapping step 312 imposes constraints in phase 0 (i.e., i mod 2=0).
x0x0x0x0x0
Step 313 imposes periodically-varying symbol substitutions during phase 1. Table 4 below illustrates the symbols that are prohibited by step 313 and the symbols that are substituted for the prohibited symbols. The symbol substitutions shown in Table 4 are performed where prohibited symbols [ci]2 in phase 1 are replaced by substitute symbols [di]2. The periodically-varying symbol substitutions are characterized by the mapping g1(.). Note that x in Table 4 stands for (xi, x2, x3, x4, x5) and ˜ stands for the Boolean negation.
x
10x20x30x40x50
All the substitutions in phase 1 are relative to the list of allowed symbols in phase 0, and therefore no substitution is performed in phase 0. The substitute symbols in phase 1 are in the set of symbols (0x0x0x0x0x) that were excluded in phase 0 at step 312.
The maximum possible rate for this class of codes with s=10 is 209/210 because 2209≦99221 and 2219>99222. The codes in this family have a symbol alphabet of size L=2s/2(2s/2−1) and satisfy the constraints G=2(s−1) and I=3(s/2)−2, where we assume that s is even and the code word size, n bits, is divisible by 2s. Although 210 is not divisible by 20, a modulation encoder of type shown in
As another example of the present invention, a rate-199/200 PRML(G=16, I=13) code with parameters n=200, L=991 and s=10 can be obtained, if the symbol 1000000000 in phase 0 and the symbol 0000000001 in phase 1 are excluded, in addition to the other excluded symbols. This code is referred to as Code 3. A modulation encoder 320 for Code 3 is shown in
x0x0x0x0x0
The range of the symbol mapping f(.) performed at step 322 is a size-991 list of 10-bit symbols that is obtained by excluding 33 symbols of type 0x0x0x0x0x and 1000000000 from the list of all 10-bit symbols. The particular mapping f(.) is preferably selected to minimize the complexity of the Boolean circuit implementing f(.). As it can be seen from Table 5, step 322 imposes constraints that are desired in phase 0 (i.e., i mod 2=0).
In order to impose the desired constraints in phase 1, the periodically-varying symbol substitutions are performed in step 323. The periodically-varying symbol substitutions performed at step 323 are shown in Table 6 below. Note that x in Table 6 stands for (x1,x2,x3,x4,x5) and ˜ stands for the Boolean negation. The symbol substitutions performed at step 323 are characterized by the mapping g1(.) in
x
10x20x30x40x50
x
10x20x30x40x50
All the substitutions in phase 1 are relative to the list of allowed symbols in phase 0, and therefore no substitution is performed in phase 0 at step 323. The maximum rate for this class of codes with s=10 is 209/210, because 2209≦99121 and 2219>99122. The codes in this family have a symbol alphabet of size L=2s/2(2s/2−1)−1 and satisfy the constraints G=2(s−1)−2 and I=3(s/2)−2, where s is assumed to be even, and the code word size, n bits, is divisible by 2s. A rate-209/210 PRML(16,13) code with parameters n=210, L=991 and s=10 code can also be constructed, although 210 is not divisible by 20. In this case, the first symbol of a code word j occurs in phase (j mod 2), whereas in
Unbalanced interleaved codes are a subclass of codes with fixed-size, periodically-varying symbol alphabets and can be implemented by simple encoding/decoding structures based on interleaving unconstrained input data with constrained data. The codes in this class have a symbol alphabet of size L=2s/2(2s/2−1) and satisfy the constraints G=2(s−1) and I=3(s/2)−2, where s is assumed to be even, and the code word size, n bits, is divisible by 2s.
Referring to
The first symbol in a word of user data is usually considered as the 0th symbol, and therefore is an even numbered symbol. The next symbol odd, the symbol after that even, and so on. The first bit in a symbol however is usually considered to be the 0th bit, and therefore is an even bit. The second bit in a symbol is odd, the bit after that even, and so on.
The 99-bit word [ai] is encoded into a 100-bit word (of constrained data) at radix conversion step 402. At step 402, the 99-bit number at the input of the encoder [ai] is converted into a radix 31 integer representation [bi]. The i index (0–98) in the [ai] integer corresponds to 99 bits of [ai]. The index i (0–19) in the [bi] integer corresponds to twenty symbols of [bi]. Each of the symbols of [bi] can be represented by 5 bits.
At step 403, a substitution operation replaces prohibited patterns in [bi]. The substitution operation 403 replaces bit patterns 00000 (bi=0) with substitute patterns 11111 (ci=31) to generate a constrained integer having 20 symbols [ci], each symbol being represented by 5 binary bits. Because the even numbered bits are extracted from the even numbered symbols and the odd numbered bits are extracted from the odd numbered symbols at step 401, substitution step 403 is able to eliminate all of the prohibited symbols (x0x0x0x0x0 and 0x0x0x0x0x) in the input data, without performing the multiple phase substitutions used in previous embodiments.
After the substitution step 403, the constrained symbols [ci] and the unconstrained 100 bits from word 420 are multiplexed together at step 404 to generate a 200 bit word 421 as shown in
According to another embodiment of the present invention, a simple and efficient algorithm based on Homer's rule can be used to implement radix conversion between 2s/2 base number representation and (2s/2—1) base number representation. A radix conversion algorithm that is based on matrix multiplication is discussed in U.S. Pat. No. 6,236,340.
For s=10, the algorithm on the encoder side for Code 2 is based on evaluating the 99-bit integer number at the input of the radix conversion engine using 19 successive multiply-and-adds in base 31 arithmetic. Because multiplication by 32=31+1 in base 31 arithmetic is equivalent to shift-and-add, only 19 additions in base 31 arithmetic are needed to obtain twenty 5-bit base 31 numbers, according to equation (1).
The radix conversion algorithm in the encoder can be implemented by 2+3+4+ . . . +20=209 successive 5-bit full adders with two 5-bit addends and 1-bit carry at its input, one 5-bit 31-base number and 1-bit carry at its output. The computational unit for 5-bit additions can be implemented using Boolean logic, where 11 bits are mapped into 6 bits and the logic circuit is optimized in terms of latency. Furthermore, two or more additions can be lumped together and further optimized for delay and gate count. For example, two 5-bit additions can be performed by a Boolean circuit mapping 21 bits into 11 bits.
The algorithm on the decoder side is based on evaluating the 100-bit integer number at the input of the radix conversion engine using 19 successive multiply-and-subtracts in base 32 arithmetic. Because multiplication by 31=32−1 in base 32 arithmetic is equivalent to shift-and-subtract, only 19 subtractions need to be performed in base 32 arithmetic to obtain 20 5-bit base 32 numbers, as shown in equation (2).
The radix conversion algorithm in the decoder can be implemented by 2+3+4+ . . . +20=209 successive 5-bit subtractions.
The constraints of the three families of PRML(G,I) codes previously discussed were characterized at the input of the Reed-Solomon (RS) encoder. The code parameters for these codes are summarized below in Table 7. As discussed above, L represents the base of the number generated at the radix conversion step, p is the number of phases, G is the global constraint, I is the interleaved constraint, and s is the number of binary bits in each symbol.
From an overall-system viewpoint, the constraints at the input of a recording channel are important. In other words, the maximum length of DC- and Nyquist patterns at the recording channel input, and the maximum length of DC-patterns in an odd or even interleave of the recording channel need to be computed. These parameters can be determined by studying the modulation properties of the sequences at the output of the partial symbol interleaver 114 shown in
The depth of partial symbol interleaving should be chosen such that the modulation constraints at the recording channel input are as tight as possible. One way that modulation constraints at the recording channel input can be tightened is by interspersing unconstrained RS parity symbols in between constrained RS data symbols, so that no 2 RS parity symbols are next to each other. Examples of this technique are described below with respect to
Table 8 below summarizes the modulation constraints following partial symbol interleaving and inner parity bit insertion for Code 1, Code 2, and Code 3, where the inner parity code is assumed to be a rate-100/102 code.
While the present invention has been described herein with reference to particular embodiments thereof, a latitude of modification, various changes, and substitutions are intended in the present invention. In some instances, features of the invention can be employed without a corresponding use of other features, without departing from the scope of the invention as set forth. Therefore, many modifications may be made to adapt a particular configuration or method disclosed, without departing from the essential scope and spirit of the present invention. It is intended that the invention not be limited to the particular embodiment disclosed, but that the invention will include all embodiments and equivalents falling within the scope of the claims.
| Number | Name | Date | Kind |
|---|---|---|---|
| 4707681 | Eggenberger et al. | Nov 1987 | A |
| 4818969 | Krakauer et al. | Apr 1989 | A |
| 5241309 | Cideciyan et al. | Aug 1993 | A |
| 5311521 | Fitingof et al. | May 1994 | A |
| 5757294 | Fisher et al. | May 1998 | A |
| 5757822 | Fisher et al. | May 1998 | A |
| 5760718 | Schouhamer Immink | Jun 1998 | A |
| 6009549 | Bliss et al. | Dec 1999 | A |
| 6018304 | Bessios | Jan 2000 | A |
| 6236340 | Patapoutian et al. | May 2001 | B1 |
| 6417788 | McEwen et al. | Jul 2002 | B1 |
| 6557124 | Cideciyan et al. | Apr 2003 | B1 |
| 6959412 | Silvus et al. | Oct 2005 | B1 |
