1. Field of the Invention
The present invention relates to a combinational circuit, and an encoder, a decoder and a semiconductor device that use this combinational circuit. More specifically, the present invention relates to a combinational circuit that can effectively correct errors, especially in a fast optical communication field, and an encoder, a decoder and a semiconductor device that use this combinational circuit.
2. Brief Description of the Prior Art
Importance of Fast and Superior Error Correction Technique
In consonance with the expansion of the Internet and the development of e-business, the rate of increase in the volume of data computers can handle and their speed has accelerated. Accordingly, there is a demand for increasing speed of data transfer among computers, and in line with this demand, optical communication that yields transfer speed of up to 40 Gbps is becoming popular. However, for such a fast communication method such as the optical communication method, to maintain an acceptable error rate at system level requires that the reliability of data communication be further increased in proportion to the amount of data processed by a computer.
Important techniques, called error correction coding techniques, have been devised to improve data reliability employing high-level mathematics to automatically correct errors caused by a variety of factors (e.g., noise along a channel). Popular known techniques are Hamming code and Reed-Solomon code, both of which are frequently employed. Basically, Hamming codes correct single bit errors, but their correction capability is low. For instance, with Hamming codes, if a single bit error is detected, the error is corrected, but if the detected error covers two bits, only the error detection portion of the process is performed, no correction is made. However, setting up an error correction process for an error correction system using Hamming code is simple, and it is well known that by performing the error correction process in parallel, a processing speed that greatly exceeds 1 Gbps (one billion bits per second) can be obtained.
On the other hand, Reed-Solomon code is a superior error correction technique possessing high correction capabilities and can be used to correct errors appearing as units (symbols) comprising multiple contiguous bits. At present, however, because of the complicated calculations that are frequently required, using Reed-Solomon code to perform correction processes in parallel is difficult; and even when pipeline processes using data having eight-bit width are performed at 100 MHz, for example, only processing speed of around 800 Mb/s can be obtained. Currently, therefore, since the speed attainable with conventional techniques using Reed-Solomon codes is not suitable for fields in which high processing speed is required, these techniques are employed mainly for fields to which comparatively low data processing speed is acceptable, such as the low speed communication field and the data storage-unit field for the production of hard disks or secondary storage devices, CD-ROMs, for example.
Error Correction Technique Required by a Fast Optical Communication Field
As part of a fast optical communication technique for data communication by computers, by the recently popular Internet as a backbone, a terabit per second fast communication system that employs WDM (Wavelength Division Multiplexing) and DWDM (Dense WDM), which has improved wavelength division multiplexing levels, has been introduced based on the SONET technique, according to which frames having a predetermined length are synchronously and sequentially transmitted.
As the wavelength division multiplexing levels for the above described optical data communication is increased, crosstalk occurs between wavelengths that are near each other. To cope with this crosstalk, FEC (Forward Error Correction) is employed as an error correction method for communication in long distance (Long Haul) optical wavelength division multiplexing. In ITU-T G.975, the ITU (International Telecommunication Union) standardized the use of interleaved (255, 239)RS code (code length n=255 bytes) of m=8 (8 bits/symbols), and in G.709, the Digital Wrapper standard for defining the FEC frame structure is employed.
According to the Digital Wrapper standard, for example, low-speed serial Reed-Solomon code circuits are arranged in parallel to achieve a necessary processing capability, and for this, interleaving of Reed-Solomon codes is an indispensable technique.
Prior Art for Fast and High-Level Error Correction Techniques
Independent of the need for optical communication, parallel fast decoding using Reed-Solomon code has been studied using a combinational circuit.
Since the decoding circuit in
When the decoding circuit in
However, the decoding circuit in
According to OC-768 SONET, this is a large problem, because assuming the 16 interleave defined by ITU-G709 is employed as an input/output interface for the decoding circuit, a fast processing speed of 300 MHz or higher is expected. Therefore, as one attempt, the decoder in
However, even when the process is converted into a pipeline, the decoding circuit in
For the calculation of error locations and error values used for the decoder, a large number of calculations in the Galois extension field GF(2m) must be performed at high speed, and further, the size of a circuit that can perform this processing must be such that it facilitates the implementation of the circuit. Conventionally, in the studies of the calculations over a Galois field, it is important that how efficiently a single calculation (multiplication or division) can be performed, and the several tens to hundreds of calculations by a combinational circuit have almost never been discussed to date. As one of various reasons this has not been done, it may be presumed many decoding operations tend to be performed by sequential circuits, and it has been ascertained that the use of a combinational circuit provides little merit in terms of processing capabilities and an acceptable circuit size.
During the studies of the error correction calculation algorithm, the Yule-Walker equation that is defined for the Galois extension field GF(2m) is generated in decoding of the Reed-Solomon codes. The efficient processing of this Yule-Walker equation is desirable if high-speed processing is to be achieved and the size of the necessary circuit is to be minimized. When the algorithm for solving the Yule-Walker equation is performed by a combinational circuit to achieve high-speed processing, in as the required error correction capabilities increase, the portion of the circuit used to solve the Yule-Walker equation and to locate errors becomes very significant from the viewpoint of the reduction in the size of the combinational circuit.
In addition, when a combinational circuit that can carry out the decoding of Reed-Solomon codes is applied for an actual system, it is preferable that an algorithm be provided that can be applied for the decoding of Reed-Solomon codes having an arbitrary minimum distance in order to obtain a process that can be widely used and to remove superfluous, additional circuits or processes. Especially in the optical communication field, since the use of (255, 239) Reed-Solomon code is standardized by the ITU, an algorithm is required that can efficiently decode the Reed-Solomon code where the maximum number of correctable errors is 8 and the minimum distance is 17.
In order to solve the mathematical problem posed by the Yule-Walker equation by using a hardware combinational circuit having a size that permits it to be implemented, increase in the circuit size must be suppressed, and an algorithm that can reduce the number of multipliers and a combinational circuit that can efficiently employ this algorithm are required. That is, a combinational circuit is needed that has an implementable size and that performs high-speed processing, and that includes the error correction device and the error correction algorithm described above.
To resolve the above shortcomings, it is one object of the present invention to provide an efficient combinational circuit for processing interleaved Reed-Solomon codes, and a signal processor and a semiconductor device that use this combinational circuit, for fast optical communication (40 Gbps or higher), or more specifically, for SONET using wavelength division multiplexing communication, in which sequential data are transferred as synchronous frames.
That is, according to the invention, a combinational circuit, which has a high processing capability (low latency and high throughput), and an encoder, a decoder and a semiconductor device that employ the combinational circuit are provided.
It is another object of the present invention to provide a flexible combinational circuit that can process interleaved Reed-Solomon codes without losing the above described characteristics, and an encoder, a decoder and a semiconductor device that employ the combinational circuit.
It is an additional object of the present invention to provide a combinational circuit that sequentially outputs error words at high speed and at a constant cycle rate, regardless of error patterns (error values and error locations) of each interleaved received word, and an encoder, a decoder and a semiconductor device that employ the combinational circuit.
It is a further object of the present invention to provide a combinational arithmetic circuit, circuits over a Galois extension field GF(2m) (m is an arbitrary natural number equal to or greater than 2), that has several common inputs and performs multiple multiplications (e.g., AB, AC and AD) and a circuit that performs the logical sums such as AB+CD+EF+ . . . can be implemented as small, fast and efficient processing circuits, and to provide an encoder, a decoder and a semiconductor device that employ the combinational circuit.
To perform the decoding of Reed-Solomon codes by a combinational circuit and to apply this combinational circuit to an actual system, it is a still further object of the present invention to provide a combinational circuit that can provide a flexible process that can be applied for the decoding of Reed-Solomon codes having an arbitrary minimum distance and interleave configurations without an additional circuit or process being required, and to provide an encoder, a decoder and a semiconductor device that employ the combinational circuit.
The above objects can be achieved by providing a combinational circuit according to the present invention, and an encoder, a decoder and a semiconductor device that employ the combinational circuit.
According to the present invention, a combinational circuit comprises:
According to the invention, a combinational circuit performing logical sums for a Galois extension field GF(2m) (m is an integer equal to or greater than 2) comprises: a plurality of multipliers, each of which includes an adder connected between an AND calculator and an output side XOR calculator, wherein the output side XOR calculator is used in common, and wherein the outputs of the AND calculators in the multipliers are added by the adders, and the addition results are calculated by the output side XOR calculator that is used in common. In the combinational circuit of this invention, the input of the multipliers is commonly used, and the input side XOR calculator is used in common by the multipliers. The combinational circuit is used for an error location calculator, for calculating error locations for a digital signal transmitted using wavelength division multiplexing, and for an error value calculator. Syndromes obtained by the coded digital signal is input. The combinational circuit of this invention is used for decoding, error correction or encryption.
Further, according to the invention, an encoder and a decoder including the combinational circuit are provided.
According to the invention, a semiconductor device used for processing a digital signal comprises:
The preferred embodiment of the present invention will now be described by referring to the accompanying drawings. Note, however, that the present invention is not limited to this embodiment.
Section 1 Decoding Circuit
In this invention, the digital signal ID input to the decoder in
In the decoder design in the invention in
As is shown in
In the decoder in
The error locator polynomial Λ(x) and the error value polynomial Er(x), which are output by the processor 12 in
In the decoder in
In the decoder of the invention in
The method or the algorithm of the invention for the error locator polynomial calculation will, along with the combinational circuit, be described in detail later. An explanation will now be given for the function and the operation of the error value polynomial calculator 20 included in the processor 12 of the decoding circuit of the invention.
Selection of an Error Value Calculation Algorithm
According to the invention, an algorithm that can directly evaluate not only error locations but also error values through the calculation of the 0(t)-degree polynomial (linear calculation) is applied for the decoding of interleaved Reed-Solomon codes. At this time, in the algorithm used for this invention, the division necessary for an error value calculation is not performed for each error location on a critical path that is output following the evaluation of a polynomial, but by using only one calculation for each code word before the evaluation of a polynomial. Therefore, the values obtained by the polynomial evaluation can be directly output as error values at high speed at a constant cycle rate.
Further, it has been found that the degree of the error value polynomial Er(x) can be reduced to the t−1 degree that is the least required to obtain t independent outputs. At this time, t coefficients can be calculated by using the coefficients of the error locator polynomial and the syndromes. By using this algorithm, not only the syndrome calculation and the error location evaluation, but also the error value evaluation can be performed merely by a linear operating circuit. Thus, the entire input/output circuit can be simplified and the processing speed can be increased.
Various error value polynomials can be used for the decoding algorithm or the decoding method of the invention. In the explanation of the function and the operation of the invention, the following example is employed wherein when e errors have occurred at io, . . . and ie−1 the error value polynomial Er(x) is provided as
and where, while the division is included, the denominator is not a polynomial but is a constant for each code word (in the above equation, a denotes the primitive element of the Galois extension field).
The error value polynomial Er(e)(x) can not be used as it is for the decoding circuit in this invention, because the k-th error value Ei
The above process can be performed by using the following decoding algorithm or decoding method. First, the denominator is written by Λj(e). The coefficients of the error value and error locator polynomials are elementary symmetric functions for error locations
ai
excluding a constant factor. For example, for coefficients of the error locator polynomial, Λ1, Λ2, . . . , Λt, the following equations are established:
The error locations al
can be obtained by a relation (e.g., see “Symmetric Functions and Orthogonal Polynomials”, I. G. Macdonald, American Mathematical Society, 1998) that is established between the Elementary symmetric function and the Shur function defined by the division of two Vandermonde matrixes, and the exchange relation, (a+b)2=(a−b)2=a2+b2, that is established between the addition and the square performed for the Galois extension field GF(2m).
When the above determinant is employed, the denominator of Er(e)(x) can be written by Λi(e). When the numerator of Er(e)(x) is calculated in the same manner, as is indicated by the following equation, the coefficients of Er(e)(x) can be written merely by Si and Λi(e), and can be calculated without using El
where Λji
As is described above, when these new relations are employed, the output unit 14 can be constituted by a linear operating circuit for the Galois extension field GF(2m), without a non-linear operating circuit being required, and the processing speed can be increased.
The following configuration is employed to constitute a fast, small coding circuit that can also cope with interleaved codes based on the above error value calculation algorithm.
(1) Use of Fast Input Unit and Output Unit (Linear Operating Circuit)
First, in consideration of the structure of the code (the number of interleaves), the bus width of an input/output interface and the number of clocks, a polynomial evaluation (linear operating) circuit, which is connected to the input side to calculate syndromes, and a polynomial evaluation (linear operating) circuit, which is connected to the output side to evaluate error locations and error values, are implemented as fast sequential circuits that employ a cyclic structure that comes from the fact that RS codes are cyclic codes, and that perform a pre-process and a post-process for (n, k) Reed-Solomon code at an arbitrary number of clocks from 1 to n. Especially effective for the invention is the fact that the above described structure is also employed for an error value evaluation portion that is a critical path for the conventional method. According to the structure of the invention, a decoding circuit can be provided as an interface that can flexibly cope not only with an input digital signal that has 255 bytes for each code word, but also with ones that have 1, 3, 5, 15, 17, 51 and 85 bytes.
Table 1 shows the relation between the width of a digital signal input to the decoding circuit and the processing clock required for this input width.
As is shown in Table 1, although the required number of clocks is increased as the width of the input digital signal is reduced, the decoder of the invention can flexibly cope with it.
In addition, in this invention, an arbitrary input/output byte width other than those in Table 1 can be selected for each code word. For example, an input/output width of eight bytes can be coped with when the code length is n=256 bytes by adding a one byte dummy at the end.
(2) Connection to a Non-Linear Operating Circuit 12
In this invention, a plurality of input units 10 and output units 14 that are constituted by sequential circuits are prepared in a number equivalent to the interleaves. Between these units, the non-linear operating circuit 12 is connected that calculates the coefficients of an error locator and error value polynomials by a multiplexer, a demultiplexer and a circuit for holding the syndromes and the coefficients of an error locator and error value polynomials. As is described above, this operating circuit 12 is constituted as a combinational circuit for a non-linear operating circuit, such as a multiplier that performs non-linear calculation. Therefore, in this invention, in the case where ITU-T G.709 (255, 239) Reed-Solomon code is obtained by 16-byte interleaving, 16 sequential circuits must be prepared before and after the operating circuit 12. In other words, in this case, multiplexing and demultiplexing must be performed at a ratio of 16:1. However, in this invention, in order to multiplex and demultiplex the coefficients of the syndrome polynomial, only a signal of 128 bits (or 136 bits at the succeeding state, depending on how Λ(x) is defined) need be processed instead of 2040 bits, so that the required numbers of multiplexers, demultiplexers and buffers can be greatly reduced.
(3) Three-Stage Pipeline Operation System That Employs the Non-Linear Operating Circuit in the Center in a Time Sharing Manner
According to the present invention, the entire decoder is operated as a three-stage pipeline formed from a linear operating circuit (syndrome calculation), a non-linear operating circuit (calculation of coefficients for error locator and error value polynomials) and a linear operating circuit (evaluation of error locations and error values), and the non-linear operating circuit of a low latency is employed in a time sharing manner by serially providing syndromes for code words for the calculation of interleaved code. Therefore, it is possible to provide efficient decoding of interleaved codes whose processing capability per circuit size is high. In the OC-768 case, for example, in order to operate the three-stage pipeline, the non-linear operating circuit in the center must complete the process for each interleaved code word at a latency of about 40 ns. When the combinational circuit is implemented by the most advanced semiconductor technique (0.18 μm or better), the above decoder can be provided as an ASIC semiconductor device.
More specifically, in the invention, fast decoder and error correction devices, the sizes of which fall within an acceptable range, can be provided by synergistic effects obtained by: (1) the use of a fast input unit 10 and a fast output unit 14, especially the use of only a linear operating circuit when calculating error locations and error values, the application of the above described algorithm for the processor 12, and the use of a configuration for a multiplier that can reduce the circuit size, and (2) the employment of the three-stage pipeline operation system by, in a time sharing manner, the processor 12, which essentially performs a non-linear calculation. In section 2, which follows, an explanation will be given for a combinational circuit that comprises multipliers included in the processor 12 of the decoder of the invention.
Section 2 Combinational Circuit
The processor 12 used for the decoder of the invention is a nonlinear circuit, specifically, a combinational circuit that uses multipliers. Unlike multipliers used for a conventional combinational circuit that has two stages, consisting of an AND calculator group and an XOR calculator group, in order to perform the multiplication of the Galois extension field GF(2m), the multiplier used for this invention has three stages, consisting of XOR gates, AND gates and XOR gates.
Configuration of a Single Parallel-Multiplier
While many studies have been made of a single multiplier, a parallel multiplier (Mastrovito Multiplier), which is constituted as a combinational circuit, not as a sequential circuit, is a field of active research. For a conventional parallel multiplication circuit (hereinafter, in this specification, referred to simply as a multiplier), there are two configuration types: the AND-XOR type and the XOR-AND-XOR type, which can be converted into each other. It should be noted, however, that the AND-XOR type is generally employed when a circuit is provided for only a single multiplication. This is because while the AND-XOR type has been well studied and various methods for obtaining a small circuit have been proposed, there is no guarantee that the circuit size for the XOR-AND-XOR type will be reduced (or may be increased), and the reduction effects can not be obtained that would compensate for the expenditure of the effort a complicated design operation would entail. The AND-XOR type and the XOR-AND-XOR type will be further described.
(1) AND-XOR Type
This is a typical method used for performing calculations in the same manner as are calculations performed with figures written down on paper, and generally a circuit of this type is employed. Specifically, the coefficients of two (m−1)th degree polynomials, which are multiplication arguments, are combined to prepare m2 partial products. This is the processing performed by the AND section. Then, partial products that have the same degree are added together to form a (2m−2)th degree polynomial, and a residue operation using an irreducible polynomial is performed to obtain the (m−1)th degree solution. This is the processing performed by the XOR section. The number of ANDs is m2 and the number of XORs is O(m3), and it is widely known that (m2−1) XORs can be obtained by selecting an appropriate irreducible polynomial and basis. An arbitrary multiplier can always be constituted by this method.
(2) XOR-AND-XOR Type
Generally, according to the Boolean algebra rule (A and B) xor (A and C)=A and (B xor C), it is possible for the XOR calculation in the residue operation unit of the AND-XOR circuit to be moved in front of the AND for use as a variable pre-processor (input side XOR calculator). Thus, the XOR-AND-XOR multiplication circuit can be obtained. For moving the XOR calculation, when an even number of the same redundant terms are added to the XOR of the residue operation unit (output side XOR calculator) by the properties A xor A=0 and B xor 0=B, many XOR calculations may be moved to the front-end as a pre-processor. Since this operation can be employed, various methods for moving the XOR are available, in addition to the simple application of the distributive law. Therefore, multiple AND-XOR types are present even with the same basis or irreducible polynomial. The number of gates in the XOR-AND-XOR type can vary, and may become either greater or smaller than that in the AND-XOR type. Another method is known whereby the number of XOR gates is systematically reduced by choosing a special basis, such as the Composite Field Multiplier that will be described below.
(3) A Method for Constructing an XOR-AND-XOR Type that can be Applied Only for a Limited Field (Composite Field Multiplier)
The composite field multiplier is a multiplier construction method that can be used only in a special case, such as where m is a composite number and the basis used for the representation of an element in a field may not be an ordinary basis (such as polynomial basis or normal basis). This method will now be described in detail. When m is a composite number, extension field GF(2m) can be constructed by two fold extensions of the field GF(2). The composite field multiplier method is a method for constructing, in accordance with the extension process, a multiplier having a recursive structure. At this time, when the product of the two values Ax+B and CX+D in GF(2m) (A, B, C and D are values in the sub-field GF(2m/2))
(Ax+B)(Cx+D)=ACx2+(((A+B)(C+D)+AC+BD)x+BD)
is employed for one quadratic extension, the number of multiplications performed in the sub-field can be reduced from four to three, and the circuit size can be reduced (KOA). At the same time, the circuit can be provided as an XOR-AND-XOR structure (the addition performed before the multiplication corresponds to the XORs arranged in front of AND). It should be noted that the use of KOA is premised on the use of the composite field multiplier, otherwise, KOA can not be used. If the value of m is the composite number for which this method can be applied, a field converter is required, and the circuit size is increased because of the overhead. Thus, when a circuit is to be prepared for only a single multiplication operation, the circuit structure does not generally, employ the composite field multiplier.
Configuration of a Combinational Circuit According to the Invention Using Ordinary Multipliers
In
Table 2 shows the numbers of gates that are included in multiplication circuits for a Galois extension field GF(28). They are conventional multiplier which has two-stage AND-XOR structure, a composite field multiplier, and a multiplication circuit, for which XOR-AND-XOR is employed for the multiplication of a sub-extension field GF(24).
AS can be seen from Table 2, as far as only a single multiplier is concerned, the circuit sizes for cases a. to c. are increased when compared with the size of the conventional composite field multiplier, and are not the minimum size. As is described above, when the three-stage XOR-AND-XOR structure is simply employed for a single multiplier, in some cases the circuit size may be increased.
Multiplier Structure for a Combinational Circuit According to the Invention
Generally, optimization of Boolean algebra is difficult when many multiplications and logical sum calculations are performed together. However, in consideration of the use of the combinational circuit as the processor 12 for the invention, many logical sum calculators and many multipliers are connected in parallel at multiple stages. The i-th logical sum calculator generally receives 0-th to (i−1)th outputs. Therefore, since the backend part of the operation circuit must process in parallel inputs used in common by almost the entire combinational circuit, the optimization range is extended. The present inventors focused on this point, and achieved an efficient configuration for a multiplier by performing the optimization for Boolean algebra, while taking into account the balance obtained with other operations.
The outputs of the AND groups 80 and 82 are transmitted to an addition circuit 84 and are added together, and the residue operation and the multiplication are again performed by a backend XOR group 86 that is used in common, so that an output 88 is generated. The structure of one multiplier in
Further, in the invention, when the XOR groups 74 to 78 on the input side, which perform the variable pre-process, are used in common, and the XOR groups that perform the residue operation are also used in common, the overall structure of the combinational circuit can be simplified.
Therefore, when, based on the multiplication circuits in Table 2, a part of the decoder is to be constituted by the combinational circuit in
It is apparent from a. to c. in Table 3, that even if the single multiplier is not intentionally minimized, the overall size of the circuit can be reduced.
In the embodiment in
Section 3 Error Correction Algorithm
A detailed explanation will now be given for an error correction algorithm that is used by the decoding circuit and the error correction device of the invention.
Overview of the Conventional Example
A. Conventional Method for Solving the Yule-Walker Equation or for Obtaining an Error Locator Polynomial, and a Problem Associated With These Methods
According to the invention, it is necessary to find an efficient algorithm for a combinational circuit to calculate the following simultaneous linear equation, which is defined over GF(2m),
where S0, S1, . . . S2t−1 are the elements of a given GF(2m), and Λi(l) is an unknown amount.
In this simultaneous linear equation, the matrix on the left has a regular structure wherein the same elements are arranged obliquely to the right (the direction irtersecting the diagonal line), and is called a Hankel matrix. Generally, this type of equation is called a Yule-Walker equation, and it is known that this equation is widely applied for various fields, such as the error correction code theory, time-series analysis and signal processing. In the error correction algorithm, the Yule-Walker equation appears in the portion for determining an error locator polynomial. Therefore, according to the present invention, the algorithm for obtaining the solution for the Yule-Walker equation is applied as an error correction algorithm for decoding the Reed-Solomon codes.
The well known methods for solving the Yule-Walker equation are, for example, the algorithm proposed by Levinson and the algorithm proposed by Levinson-Durbin. These algorithms start calculating with a matrix of the smallest size l and recursively determine the solution of an equation wherein the size of a matrix is greater. The number of calculations required by these two algorithms is of order of l2. However, these algorithms include divisions in the calculation step. This means that when the algorithm is mounted as a combinational circuit, conditional branches occur depending on whether the denominator is 0 or not. Since a separate circuit must be prepared for each conditional branch, the required circuit size increases as the size of the matrix increases.
Further, as the object of the invention, especially relative to the decoding of the Reed-Solomon codes, an error locator polynomial is determined by obtaining the solution to the Yule-Walker equation. The conventional methods for solving the Yule-Walker equation can be, for example, the Peterson method, the Berlekamp-Massey method and the Euclid method. These methods are used to calculate the coefficients of an error locator polynomial by calculations of which the number is polynomial order with respect to the maximum number of correctable errors t. However, when the Berlekamp-Massey method and the Euclid method are represented by a combinational circuit, the following problems occur.
First, for the Berlekamp-Massey method, it is inevitable that multiple conditional branches should be included in the algorithm. Therefore, to expand this algorithm for a combinational circuit, for the same reason as described above, the circuit size would be increased in accordance with the number of combinations. As for the Euclid method, while the multiplication and division of the polynomial are the essential part of the algorithm and, the degree of a polynomial that appears in the denominator of the division can not be identified in advance, so that there is room for generating a conditional branch. Furthermore, clue to the conditional branches, the circuit size is accordingly increased, as it is for the Berlekamp-Massey method.
B. Policy for Calculating the Yule-Walker Equation and an Error Locator Polynomial that is Appropriate for a Combinational Circuit
Since, as is described above, the Levinson(-Durbin) method, the Berlekamp-Massey method and the Euclid method include conditional branches, a problem has arisen in how to provide these methods as combinational circuits. In order to implement the Yule-Walker equation by a combinational circuit, an algorithm that has no conditional branching must be found, and this is an essential object for the algorithm of the invention.
In this case, the Peterson method known for decoding of the Reed-Solomon codes can be used as the algorithm of the invention. Using the Peterson method, the Yule-Walker equation can be solved directly, and the solution of the Yule-Walker equation can be represented as determinants by the Cramer formula:
Therefore, determinants ΛO(l) need be obtained for each l=1, . . . , t, and the determinants {tilde over (Λ)}i(e), i=1, . . . , e need be calculated for the number of errors e.
However, when the expansion of the determinants is provided by using a circuit, the required number of multipliers is dramatically increased as t is increased, so that it is difficult for the determinants to be directly expanded. Therefore, in this invention, the number of calculations is reduced by the recursive structure of the Hankel matrix. The calculation of Λinat(l) using the Katayama-Morioka method will be explained.
When the calculation algorithm for Λi(l) in the Katayama-Morioka method is written for l=1 to l=4, the form in
For comparison, a method devised by Koga will now be described as another method for recursively calculating the Hankel matrix. According to the method by Koga, new error locator polynomial i;l+2uD(X, Y) is defined, wherein
To calculate this new error locator polynomial, the Hankel matrix i;l+2uQ, which has the i-th syndrome Sl as the (1, 1) element, is employed, and a determinant obtained by symmetrically removing multiple rows and columns from this Hankel matrix is defined as a Q determinant. When the subscript numbers of syndromes that appear as a diagonal element are designated in order beginning at the upper left, only one Q determinant can be determined. In this case, the Q determinant can be represented by the row of subscripts [a1, a2, . . . , ap]. According to the Koga method, an algorithm is presented by to calculate error locator polynomial i;i+2uD(X, Y) the Q determinant.
All of the above described conventional methods have the following problems. First, for the algorithm used in
As for the Koga algorithm, the Q determinant defined by Koga is symmetrical, and a reduction in the number of multipliers is carried out. However, there is a limitation on the use of the Koga algorithm; this algorithm can be applied for BCH codes or Reed-Solomon codes only when the minimum distance is an even number. Although it is disclosed in the Koga algorithm that this limitation can be eased, such an application example is limited to the binary, narrow sense BCH code.
According to the invention, since the combinational circuit is applied for an optical communication system, it is required as an object that the decoding of the (255, 239) Reed-Solomon code (minimum distance=17) be efficiently carried out by the combinational circuit. Therefore, an algorithm is needed that can perform an efficient calculation by a specific method, regardless of whether the minimum distance is an odd or even number.
C. Definitions of Terms Used for this Invention
Before a detailed explanation is given for the algorithm of this invention, definitions for the terms used for this invention will be given.
(1) Syndrome
Generally, when a primitive element of the Galois extension field GF(2m) is defined as a, and h<2m−1 is a positive integer, a 2m-element cyclic code that has a code length of n=2m−1 and that employs
G(x)=(x−1)(x−a)(x−a2) . . . (x−ah−1)
as a generator polynomial is defined as the Reed-Solomon codes. That is, when k=n−h and when the k-th degree polynomial having k information symbols as its coefficients is defined as M(x), M(x) and xn−k are multiplied, and the result is divided by G(x), as follows, to obtain the residue R(x).
M(x)xn−k=Q(x)G(x)+R(x)
Then, the polynomial (transmission polynomial) that has as coefficients a coded sequence having the length n is defined as
W(x)=M(x)xn−k−R(x)=Q(x)G(x).
At this time, the coded transmission sequence is represented as systematic code, of which k information symbols are located on the left and h=n−k check symbols follow these symbols. The minimum distance dmin=h+1 of the Reed-Solomon code, and the maximum number of correctable errors t=[h/2] are provided.
The following decoding algorithm is given for making an estimation of the original transmission sequence based on a received sequence.
(2) Calculation of Syndromes and Detection of an Error
Assume that errors have occurred, and that the locations of the errors are denoted by io, . . . , il−1 and the error values are denoted by Ei
E(x)=Ei
a polynomial having a received sequence of b0, . . . , bn−1 as coefficients is provided as
where Y(x) is defined as a reception polynomial.
Then, the reception polynomial Y(x) is employed to calculate the following syndromes:
Si=Y(ai)εGF(2m), i=0, 1, . . . , 2t−1.
In this case, since W(ai)=0, i=0, 1, . . . , 2t−1 is established, and the obtained syndromes satisfy Si=E(ai), i=0, 1, . . . , 2t−1. Therefore, when there is no error, all the syndromes are 0, so that the value of syndromes can be used to determine the presence of errors.
(3) Determination of the Number of Errors and the Location of Errors
Assume that the number of errors that occur is 1 and that the location of the error is i0, . . . , il−1, i.e., assume that the values of bl
where a−l
Further, Λ1(l), . . . , Λl(l) are coefficients of the error locator polynomial and are provided using the elementary symmetric function of ai
The unknown quantities Λ1(l), . . . , Λl(l) satisfy the following simultaneous linear equations:
This is nothing but the Yule-Walker equation explained in section A. While at this step “1 ” is an unknown number, it is known that when the number of errors that actually occurred is 1≦e≦t, the Hankel matrix on the left is regular when l=e and is irregular when t≧l>e. Therefore, for l=1, . . . , t, only the determinants for the Hankel matrix on the left need be calculated, and a maximum integer that is not 0 can be defined as the number of errors e. When the above equation is solved with 1=e, the error locator polynomial can be obtained.
In this invention, to specify error locations, the error locators, i.e., the roots of error locator polynomial Λ(e)(x)=0, need be calculated. For this calculation, a method can be employed whereby a−i, i=0, 1, . . . , n−1, is substituted in to determine whether the result is the zero point of the error locator polynomial. This method is called the Chien search method. When the zero point of the error locator polynomial is denoted as a−i
4. Calculation of Error Values
Error values can be obtained by solving the following Vandermonde simultaneous linear equations:
In this case, polynomial S(x), having a syndrome as a coefficient, is defined as
S(x)=S0+S1x+S2t−1x2t−1,
and
Ω(x)=Λ(e)(x)S(x)mod x2t−1
is defined, wherein Ω(x) is called an error evaluator polynomial. In this case, the solution for the Vandermonde simultaneous linear equation can be obtained by calculating
This is called the Forney algorithm. When the error locations and the error values are obtained, only these need be subtracted from an input digital signal, so that a digital signal for which errors were corrected can be output.
D. Algorithm to Solve Yule-Walker Equation of this Invention
The object of the present inventors is to find an efficient algorithm that employs the combinational circuit to obtain the solution for the following Yule-Walker equation, defined over GF(2m):
where S0, S1, . . . , S2t−1 are the elements of a given GF(2m), and Λi(1) are unknown amounts.
In this invention, by the Cramer formula the solution for the Yule-Walker equation is represented as the determinant form shown in
In order to calculate determinants in
Jacobi's Formula
A=(aij) is defined as an order n square matrix on a commutative ring having a unit element of 1, and the cofactor (i, j) of A is defined as Δij. When Δμν(r) is defined as the cofactor of minor Aμν(r), with a set of subscripts,
μ={i1, . . . , ir, (i1< . . . <ir)}, ν={j1, . . . , jr, (j1< . . . <jr)},
the following equation is established.
In this invention, the following equation
Δi
can be employed, which holds when
μ={i1, i2}, ν={j1, j2}.
The calculation of Λihat(l) will now be described using Jacobi's formula.
First, Λihat(l+1) is represented in the following form.
By careful examination of this determinant, Λlhat(l) is obtained by removing the (l+1−i)th row and the l-th column from Λihat(l+1), and Λ0 hat(l) is obtained by removing the l-th row and the l-th column. That is, since Λ0hat(l) and Λihat(1) are (l+1, l+1) and (l+1, l+1−i) cofactors of Λ0(l+1) respectively,
Δl+1, l+1={tilde over (Λ)}0(l), Δl+1, l=1−i=Δl+1−i, l+1={tilde over (Λ)}i(l),
while i1=j1=l+1−i, i2=j2=l+1 is set in Jacobi's formula. Further, the (l+1−i, l+1−i) cofactor of Λ0 hat(l+1) is defined as Γl(l+1), the structure of which is shown in
Γi(l+1){tilde over (Λ)}0(l)+({tilde over (Λ)}i(l))2={tilde over (Λ)}0(l+1)Γi−1(l), i=1, . . . , l.
When Jacobi's formula is employed, the calculation of Λihat(l) results in the calculation of Γi(l+1), which is the determinants of symmetric matrices. It should be noted, however, that, to obtain Λlhat(l), not only the calculation of Γi(l+1) but also the 2×l multiplications and the calculation of l square roots are required. Since the calculations for the square root and the square calculations can be performed as linear calculations for GF(2m), these calculations can be implemented as a circuit substantially at the similar cost as that of an addition. Therefore, only a very small cost is required, compared with a multiplier that is a non-liner operating circuit. Therefore, the present inventors focused on only the multipliers, and discussed the number of them that would be required. Since the characteristic of the GF(2m) is always 2 and all Γi(l) are symmetrical, the algorithm proposed here always cancels terms that are generated from arrangements that are asymmetrical to the diagonal line in the process for expanding the cofactor of determinants. For example, when the cofactor expansion is calculated for a 3×3 symmetric matrix, the following equation is obtained
Since the term “bec”, which is generated from the arrangement asymmetrical to the diagonal line, always appears twice, this term is canceled. Thus, when the algorithm of the invention is used for a combinational circuit including multipliers, the required number of multipliers can be reduced.
The general form of the algorithm for the recursive calculation of Γi(l), l=1, 2, . . . , t+1, i=0, 1, . . . , t, is provided as follows.
0. Γ1(1)=1, Γ0(2)=S0, Γ1(2)=S2
1. when 1>2, i=1,
2. when 1>2, i=1, . . . , l−1, first, one auxiliary amount for describing the algorithm is defined.
When {i1, . . . , in} is defined as a set of subscripts, det [{i1, . . . , in}] is defined as
Specifically, det [{i1, . . . , in}] is the determinant of a symmetric matrix wherein the first row is Si
In this equation, det [{0, 1, . . . , 1−2}−{l-1-i, l-1-k}] is the determinant for a symmetric matrix that is obtained by symmetrically removing, from Γi(l−1), l-1-i and l-1-k rows and l-1-i and l-1-k columns. Note that, when k=1 and i=1, the determinant matches expression of Γl−2(l−2), Γk−2(l−2).
3. Generally, det[{i1, . . . , in}] is calculated as follows.
E. Application of the Algorithm of the Invention for the Decoding of the Reed-Solomon Code
An explanation will now be given for the embodiment wherein the algorithm of the invention for solving the Yule-Walker equation described in D is applied for the Reed-Solomon codes. Generally, it is assumed that the order of the Yule-Walker equation (the number of unknown quantities) is known. However, for the decoding of the Reed-Solomon codes, since the order is also unknown, this must also be determined.
(1) Calculation of Γl(l)
When a sequence of syndromes, S0, S1, . . . , S2t−1, is provided,
Γi(l), l=1, 2, . . . , t+1, i=0, . . . , t
is calculated in accordance with the algorithm explained in
D. During this Calculation,
{tilde over (Λ)}0(l)=Γ0(i+1), l=1, . . . , t
is also calculated. It should be noted that for fast decoding of the Reed-Solomon code, the present inventors have taken into consideration the fact that the algorithm is implemented as a combinational circuit. However, use of the error correction algorithm of the invention is not limited to the combinational circuit; it can be employed as an error correction device by a sequential circuit.
(2) Determination of the Number of Errors
Assume that the number of errors that actually occurred is represented as e. Based on the value of
{tilde over (Λ)}0(l)=Γ0(l+1), l=1, . . . , t,
e can be obtained as the maximum “l” that satisfies Λ0hat(l)≠0.
(3) Determination of an Error Locator Polynomial
When e<t is established as the result of the determination of the number of errors, since Λ0hat(e+1)32 0, in accordance with the algorithm of the invention, the above equation can be simplified as follows:
{tilde over (Λ)}i(e)=√{square root over (Γi(e+1){tilde over (Λ)})}{square root over (Γi(e+1){tilde over (Λ)})}0(e), i=1, . . . , e.
Since the error locator is the zero point of the error locator polynomial, the error locators are values unchanged by multiplication of the coefficients of the error locator polynomial by a constant. Therefore, the following quantity
√{square root over (Γi(e+1))}
can be used instead of Λihat(e). In other words, the multiplication appearing in the above equation is not required. When e=t is established, the error locator polynomial is calculated in accordance with the following equation:
{tilde over (Λ)}i(e)=√{square root over (Γi(e+1){tilde over (Λ)})}{square root over (Γi(e+1){tilde over (Λ)})}0(e)+{tilde over (Λ)}0(e+1)Γi−1(e), i=1, . . . , e.
At this time, according to the algorithm proposed by the present inventors, in appearance, syndrome S2t, which can not be calculated, seems to be necessary when the minimum distance is an odd number (=2t+1). However, since the equation of this invention serves as the identity of the syndromes, it can also serve as the identify of the syndrome S2t. Furthermore, since Λihat(t) does not include syndrome S2t, when syndrome S2t appears during the cofactor expansion for Λ0hat(t+1) and Γl(t+1), it should always be canceled. Specifically, since a term that includes S2t and that appears during the cofactor expansion of Γi(t+1) is Γi−1(t)S2t, when Γl(t+1)Λ0hat(t) is expanded, a term including S2t is Λ0hat(t)Λi−1(t)S2t. Further, since, of the terms that appear during the cofactor expansion of Λ0hat(t+1), Λ0hat(t)S2t includes S2t, a term including S2t when Λ0(t+1)Γi−1(t) is expanded is Λ0hat(t)Γi−1(t)S2t. Therefore, all these terms must always be canceled.
It is therefore understood that, of the terms that appear during the cofactor expansion of Λ0(t+1) and Γi(t+1), terms having S2t as a coefficient need not be calculated. In this manner, the algorithm of the invention can be applied for Reed-Solomon codes having an arbitrary minimum distance. Further, since the multiplication of the term including S2t is not necessary, from the viewpoint of the reduction in the number of multipliers, the algorithm of the invention is superior to the Koga algorithm. In addition, as is described above, the calculation of a square root can be implemented as a circuit at the same cost as that for addition, and only a small cost is required compared with the cost of a multiplier.
F. Example of Application of the Error Correction Algorithm of the Invention for the Decoding of Reed-Solomon Codes
An explanation will now be given for a case wherein the error correction algorithm explained in E. is employed for the decoding of Reed-Solomon code for t=4. When t=4, according to the invention the following equations are determined. It should be noted that for simplification the determinant is represented as det [{i1, . . . , in}]=deti1. . . in.
(1) Calculation of Γi(l), i=0, . . . , l−1, . . . , 5
The calculation results obtained by the invention are shown in
(2) Determination of the Number of Errors
Since {tilde over (Λ)}0(l)=Γ0(l+1), l=1, . . . , 4 is determined by Γi(l), obtained in (1), and the number of errors e can be determined as the maximum 1, l=1, 2, 3, 4, that satisfies
{tilde over (Λ)}0(l)≠0.
(3) Determination of an Error Locator Polynomial
When, for example, e=2 is ascertained using the calculation in (2), the error locator ai
√{square root over ({tilde over (Λ)}0(2)+√{square root over (Γ1(3))}χ+√{square root over (Γ2 (3))}χ2=0
When e=4 is ascertained, as is described above, the error locator can be obtained by
{tilde over (Λ)}i(4)=√{square root over (Γi(5){tilde over (Λ)})}{square root over (Γi(5){tilde over (Λ)})}0(4)+{tilde over (Λ)}0(5)Γi−1(4), i=1, 2, 3, 4
It should be noted that, as is described above, the term including the syndrome S8 need not be calculated when the Γi(5), Γ0(6)={tilde over (Λ)}0(5) calculations are performed.
G. Calculation Circuit when the Algorithm of the Invention is Applied for the Calculation of an Error Locator Polynomial
The functions of the blocks in
Following this, the circuit block 102 employs the obtained value Γ0(m), m=1, 2, . . . , t+1 to calculate the number of errors e, and outputs Γi(e+1), i=0, . . . , e, which corresponds to the value of e. When e=t, in addition to the above, Γ0(t+2)={tilde over (Λ)}0(t+1) is also output. This corresponds to detailed explanation (2) for the algorithm. The circuit block 104 then employs Γi(e+1), i=0, . . . , e to calculate the coefficients of the error locator polynomial. This calculation is performed in accordance with the process corresponding to detailed explanation (3) for the algorithm.
The algorithm of the invention has been used for the combinational circuit in order to perform fast decoding of the Reed-Solomon codes. However, the algorithm of the invention can also be used for a sequential circuit in order to reduce the circuit size.
H. Circuit Size when the Algorithm of the Invention is Used for the Decoding of the Reed-Solomon Codes
An explanation will now be given for the size of a circuit when the algorithm of the present invention is used for the decoding of the Reed-Solomon codes. As is described above, the calculation of square roots and the calculation of squares can be performed by a circuit having substantially the same cost as an addition circuit, and compared with a multiplier, the cost required is very small. The present inventors have focused only on the multipliers, and discussed the number of multipliers that are required.
Table 4 shows the number of multipliers required by the algorithm of the invention in a range extending from t=1 to t=8. In Table 4, for comparison, the number of multipliers required for each of the conventional examples 1 and 2 is also shown.
As is apparent from Table 4, while taking the required number of multipliers into account, the algorithm proposed in this invention is superior in all number of errors t to the algorithm (conventional example 2) proposed by Koga. Further, the use of an algorithm for the decoding of (255, 239) Reed-Solomon code (t=8) is especially important for the optical communication field; however, the Koga algorithm can not be so employed because the minimum distance of Reed-Solomon code is an odd number (=17). Since the algorithm of the invention can be used for Reed-Solomon codes having an arbitrary minimum distance, it can also be used for (255, 239) Reed-Solomon code. This is shown in Table 5.
The calculation algorithm in conventional example 1 (Katayama-Morioka) can also be used for Reed-Solomon code having an arbitrary minimum distance. However, from the viewpoint of the required number of multipliers into account, when t is equal to or greater than 4, the algorithm proposed in this invention requires a smaller number of multipliers than does the algorithm of conventional example 1. It has especially been found that when t=8, the algorithm of the invention can reduce the number of multipliers by about 50%. And as for a circuit size, when t=8, 10K gates are currently required for the calculation of the error values. For conventional example 1 about 80K gates seem to be required, while the employment of the algorithm of the invention can reduce the gates for the calculation of an error polynomial to about 40K gates.
The process block 114 includes a decoding function employing the algorithm of the invention, and calculates error locations and error values. The error locations and the error values that are obtained are transmitted to the output block 116, the error is corrected, and the resultant digital signal is output. The above described error correction circuit can be provided as an error correction device comprising multiple hardware components, or a semiconductor technique may be employed to provide a semiconductor device, such as an ASIC, for which the individual functional blocks of the error correction circuit are implemented on a silicon wafer. In addition, the algorithm of the invention can be mounted as firmware for the error correction device, or may be provided as a computer-readable program that is recorded on a storage medium, such as a floppy disk, a hard disk, an optical disk or a magneto-optical disk. The program of the invention may be written in an arbitrary object-oriented language or a programming language such as C, and stored on the above mentioned storage medium.
As is described above, according to the present invention, it is possible to provide a combinational circuit that can extremely efficiently correct errors in the fast optical communication field, and an encoder, a decoder and a semiconductor device that employ this combinational circuit.
Number | Date | Country | Kind |
---|---|---|---|
2001-066573 | Mar 2001 | JP | national |
2001-196027 | Jun 2001 | JP | national |
Number | Name | Date | Kind |
---|---|---|---|
5107503 | Riggle et al. | Apr 1992 | A |
6327654 | Oowaki et al. | Dec 2001 | B1 |
6718138 | Sugawara | Apr 2004 | B1 |
Number | Date | Country | |
---|---|---|---|
20030063554 A1 | Apr 2003 | US |