Error counting mechanism

[0002] The invention relates generally to error correction code decoding mechanisms and more particularly to the decoding of Bose-Chaudhuri-Hocquenghem (BCH) correction codes, including Reed-Solomon error correction codes.

[0003] The use of increasingly higher density storage media in digital computer systems has caused an increase in the potential for defect-related data errors. To reduce data loss as a result of such data corruption, error correction codes are employed to correct the erroneous data.

[0004] Prior to storing data on a storage device, such as a magnetic disk or tape, it is encoded to form redundancy symbols. The redundancy symbols are appended to the data symbols to form code words, which are then stored on the storage device. When the stored data is retrieved from the storage device for decoding, the redundancy symbols provide information which allows the decoder to recognize errors and, if possible, reconstruct the original code word. A detailed description of decoding may be found in “Error-Correcting Codes,” Second Edition, by W. Wesley Peterson and E. J. Weldon, Jr. (MIT Press, 1972), incorporated herein by reference.

[0005] One widely-used error correction code is the Reed-Solomon code. Error detection and correction techniques for Reed-Solomon codes are well known.

[0006] To correct errors, a decoder must determine the locations and values (or magnitudes) of the detected errors. The decoder first computes error syndromes, which it then uses to generate an error locator polynomial. Once the error locator polynomial has been generated, each error location and value may be determined.

[0007] Error locations are determined by solving for the roots of the error locator polynomial σ(x) of degree t or less, where t is the number of errors that can be corrected. The solution or roots of the equation σ(x)=0 correspond to the locations of the errors. These roots are of the form x=αⁱ, where α is the primitive element of the Galois Field GF(p^q) used to encode the data. Once all t roots have been found, the corresponding error values are calculated using the well-known Forney algorithm. The data can then be corrected to produce an error-free data symbol.

[0008] Typically, when a codeword is read from the storage media, error correction is performed, as the corrected data is to be provided to a user. There are, however, some instances in which it is useful to determine the number of errors that have occurred in a codeword, but decoding is unnecessary. For example, in a disk test application, codewords are written to and read from a new disk and the number of errors compared to a threshold in order to assess the quality of a new disk. In yet another example, in read-after-write operations of a tape drive, which is typically less reliable than disk, it is desirable to determine the error count immediately after the completion of the read-after-write operation is completed so that the read codeword can be relocated to another storage location on that tape drive if the error count is too high.

[0009] There are several prior approaches to determining the number of errors in a codeword retrieved from storage. One approach is to use a conventional decoding algorithm to find the error locator polynomial, the degree of which corresponds to the number of errors. This approach is quite complex, as it requires the full implementation of either of the well-known Berlekamp-Massey or Euclidean algorithms, which are described in “Theory and Practice of Error Control Codes” by Richard E. Blahut (Addison-Wesley, 1983), incorporated herein by reference.

[0010] Another prior technique uses a theorem, or more specifically, Theorem 9.9 on page 284 of the above-referenced text by Peterson and Weldon. This theorem states that a v×v matrix M with syndromes as elements is singular (i.e., the determinant is zero) if there are fewer than v errors per code block, and the matrix is nonsingular (i.e., the determinant is nonzero) if there are exactly v errors per code block.

[0011] In one aspect of the invention, determining a number of errors in an error correction codeword includes determining from the error correction codeword a syndrome polynomial which is associated with an error locator polynomial of a degree and having error locator polynomial coefficients, operating on the syndrome polynomial to determine the degree without determining the error locator polynomial coefficients, and identifying the degree as the number of errors in the error correction codeword.

[0012] Embodiments of the invention may include one or more of the following features.

[0013] It is determined whether or not the number of errors exceeds an error threshold. If the codeword is stored in a storage location on a storage device and if it is determined that the number of errors exceeds an error threshold, the error correction code is rewritten to the storage location, or a different storage location, on the storage device.

[0014] In one embodiment, the syndrome polynomial operation includes a Euclidean process that determines the degree without determining the coefficients. For a t error correcting code, the performance of the Euclidean process includes initializing a divisor value as corresponding to the computed syndrome polynomial and a dividend value which corresponds to a polynomial x^2t. The Euclidean process includes the following: initializing a degree count to zero and a degree count indicator value to zero; producing a quotient value from a leading term of the dividend value and a leading term of the divisor value; generating a remainder value from the quotient value; updating the dividend for the remainder value; incrementing the degree count by one and setting the degree count indicator value to one if the degree count indicator value is set to zero. The steps of producing, generating, incrementing and setting are repeated until the difference of the degree of the divisor value subtracted from the dividend value is less than zero. The process further includes: setting the degree count indicator value to zero; interchanging the divisor value and remainder value; and determining if the degree of the remainder value is less than t. If the degree of the remainder value is less that t, then the process indicates that the degree count is the number of errors. If the degree of the remainder value is not less than. t, then the process repeats the steps of producing, generating, incrementing, setting, repeating, interchanging and determining.

[0015] In another embodiment, the syndrome polynomial includes syndrome values and the operation on the syndrome polynomial includes arranging the computed syndrome values to form a matrix having rows of unequal lengths and examining at least one row for nonzero values. The syndrome polynomial operation further includes performing row operations on the rows of the matrix. For a matrix which includes t rows, the syndrome polynomial operation includes indicating that there are fewer than t errors in the codeword if an examined last row t−1 has no nonzero values. Alternately, the step of examining is performed prior to each of the row operations. The steps of examining and performing are repeated until an examined row is equal to all zero values to indicate a number of errors corresponding to the number of the row or the row is row t to indicate that the number of errors is at least t.

[0016] In another aspect of the invention, the error number determination is used to test storage devices. Testing the storage device includes reading codewords from the storage device, generating syndrome values for the codewords, determining from the syndrome values for each of the codewords a degree of an error locator polynomial without computing the coefficients of the error locator polynomial, the degree corresponding to a number of errors. The number of errors is then compared to an error threshold. The storage device is rejected if the number of errors of a predetermined number of one or more of the read codewords exceeds threshold.

[0017] Among the advantages of the present invention are the following. The error count can be determined without computing the error locator polynomial. Therefore, the error count for a codeword is realized with a much smaller number of operations than are required by either of the Berlekamp-Massey or Euclidean algorithms. This is quite useful for applications that need to quickly establish the condition of the codeword. Because only the degree is determined, a Euclidean approach is advantageously modified to eliminate error locator polynomial computation steps, e.g., coefficient multiplications and error locator polynomial updates, which consume memory and computation cycles. Only the degree of the error locator polynomial need be updated and retained. The actual error locator polynomial, which can be obtained from all of the quotient values, need not ever be computed, and the quotient values may be discarded once the degrees of the quotients are used to update the degree of the error locator polynomial.

[0018] Additionally, the Euclidean computation polynomials have the property that degree of the remainder value (remainder polynomial) is less than the degree of the divisor value (divisor polynomial), the degree of each quotient value is at least one and, consequently, the degree of the error locator polynomial is always increasing during the iterative Euclidean process. Therefore, if after t iterations the degree of the error locator polynomial is at least t and the degree of the remainder polynomial is still not lower than t, there is no need to proceed further, since the error locator polynomial degree will exceed t for any additional computations.

[0019] The matrix evaluation makes the above-mentioned textbook theorem readily applicable to error counting by providing a condition for the case of more than v errors. The evaluation need not examine matrices of different sizes to provide an accurate error count.

[0020] Other features and advantages of the invention will be apparent from the following detailed description and from the claims.

[0021] FIG. 1 is an block diagram of a Reed Solomon decoder having an error locator polynomial degree determiner as well as a conventional error locator polynomial generator.

[0022] FIG. 2 is a flow diagram of a method of generating the error locator polynomial (as performed by the error locator polynomial generator of FIG. 1) by applying a conventional Euclidean algorithm.

[0023] FIG. 3 is a flow diagram of a method of determining a number of errors in a codeword using a Euclidean process.

[0024] FIG. 4 is a flow diagram of an alterative method of determining the number of errors using a syndrome “matrix” evaluation.

[0025] FIG. 5 is a flow diagram of a row operation of the syndrome matrix evaluation method of FIG. 4.

[0026] FIG. 6 is a block diagram of a data storage system which employs the error locator polynomial degree determiner of FIGS. 1, 3 and 4.

[0027] FIG. 7 is a block diagram of a storage device test unit which uses the error locator polynomial determiner of FIGS. 1, 3 and 4.

[0028] The present invention features a mechanism which determines the number of errors that have occurred in a codeword. The determination finds the degree of an error locator polynomial associated with the codeword's syndrome polynomial without determining the error locator polynomial coefficients. The degree of the error locator polynomial is identified as the number of errors. In one embodiment, the degree is determined by a Euclidean process that is a modification of the conventional Euclidean algorithm. In another embodiment, the degree is determined by a rank evaluation of a specially shaped “matrix” of syndrome values. Both of these embodiments will now be described.

[0029] Referring to FIG. 1, a Reed-Solomon code decoder 10 is shown. The Reed Solomon decoder 10 includes an arrangement of functional units, including a syndrome generator 12, an error locator polynomial generator 14, an error location computation (or root finding) circuit 16, an error value computation unit 18 and an error corrector 20. Also included is a memory 22 for storing a copy of a received codeword received on a data bus 24. The memory 22 stores the received codeword while the locations and values of detected errors of the received codeword are being computed by units 12, 14, 16 and 18, collectively. The units 12, 14, 16, 18 and 20 are conventional decoder blocks, the details of which are largely omitted herein. The decoder 10 also includes an error locator polynomial determiner 26 for producing an error number output value 27, which will be described in detail below. Control of each of these functional units is effected by a controller 28. Control signals are provided to the functional units by the controller 28 over a control bus 30. The syndrome generator 12, the error locator polynomial determiner 26 and the controller 28 are referred to collectively as an error count unit 31.

[0030] The decoder 10 receives as an input a codeword c(x), which may contain some error e(x). Therefore, the received codeword, i.e., c(x)+e(x), is referred to as c′(x). The codeword c(x) of n symbols includes k information or data symbols d(x) and r (i.e., “n−k”) redundancy symbols r(x). Each of the symbols is m bits in length. The redundancy symbols are derived by dividing the information symbols by a generator polynomial g(x). Division results in a quotient q(x), which is ignored, and a remainder r(x). The remainder includes the redundancy symbols which are then added to the n−k lowest order locations of c(x). With Reed-Solomon codes, the number of redundancy symbols r to correct an error must be at least twice the number of errors t that are to be corrected; thus, there must be r≧2t redundancy symbols in a t-error-correcting code. As the order of the remainder corresponds to the order of the divisor, the generator polynomial implemented for a t error correcting Reed-Solomon code having n≦2^m−1, m-bit symbols is as follows:

[0031] g(x)=(x−α^L)(x−α^L+1) . . . (x−α^L+r−1)  Eq. (1)

[0032] where αⁱfor I=L, L+1, . . . L+r−1, is a field element of GF(2^m) and L is a constant of the Galois field factor (x−αⁱ). The codeword c(x) is a multiple of both the generator polynomial and each of its factors or roots. Thus, the division of the received word c′(x) by the generator polynomial g(x) or each of its roots, if the received worked contains no errors, results in some quotient and a remainder of zero.

[0033] Still referring to FIG. 1, the codeword c′(x) is applied to the syndrome generator 12 over an input bus 32 from a storage device (not shown). For error correction purposes, it is necessary to generate a number of syndromes equal to twice the number of errors being corrected, i.e., 2t. The syndrome is defined as the remainder after dividing the received word c′(x) by a root (x−αⁱ) of the generator polynomial g(x). This is the equivalent of evaluating the received polynomial c′(x) at αⁱ, i.e., c′(αⁱ). As there are t such roots, there are 2t syndromes. The syndromes are mathematically related to error locations and error values by the relationship: $\begin{array}{cc}{S}_j=\sum _c^{t-1}{Y}_i{X}_i^j& \text{Eq. (2)}\end{array}$

[0034] where X_iare the error locations and the Y_iare the error values, and S_j=c′(α^j). X_iis an element of the Galois field and the power of α corresponds to the location of the symbol error, i.e., if X_i=α⁸³, the 83rd symbol is in error. The value of Y_iis also an element of Galois Field GF(2^m) and corresponds to the error pattern.

[0035] Therefore, the received code word c′(x) is tested for the presence of errors by computing syndromes S_j=c′(αⁱ), where j=L, L+1, . . . , L+r−1. If all syndromes S_jare zero, then there is no error and therefore no need for further decoding. If at least one S_jis nonzero, the codeword c′(x) must be subject to further processing by the decoder 10.

[0036] During a decoding process, the controller 28 directs the syndrome generator 12 to provide the computed syndromes to the error locator polynomial generator 14, which calculates the coefficients of an error locator polynomial from the computed syndromes according to known algorithms, such as the Berlekamp-Massey or Euclidean algorithms. The error locator polynomial is of the form:

[0037] σ(x)=σ_tx^t+σ_t−1x^t−1+σ_t−2x^t−2+σ_t−3x^t−3+σ₂x²+σ₁x+σ₀  Eq. (3)

[0038] for a t-error correcting code.

[0039] For a minimum distance D code having t≦D/2, the mathematical relationship between S(x) and σ(x) is well-defined by known equations that enable σ(x) to be found for a given S(x). Therefore, it can be said the syndrome polynomial generated for a codeword is associated with or implies the particular error locator polynomial from which the locations of the errors in the codeword may be obtained.

[0040] The coefficients of the error locator polynomial σ(x) generated by the error locator polynomial generator 14 are transferred to the error location computation circuit 16. The error location computation circuit 16 evaluates σ(x) for each power of α^j, j=0, 1, 2, 3, . . . , n−1 (n being the total number of symbols in the codeword c′(x)) and determines if the result is zero or non-zero. The values of σ(x) equal to zero correspond to error locations X_i, which are provided to the error values computation unit 18 and the error corrector 20, as shown. A detailed example of the error locator polynomial generator 14 is described below with reference to FIG. 2.

[0041] Still referring to FIG. 1, the error values computation unit 18 receives as inputs the error locations X_iand the syndromes S_iand computes error values Y_iusing known techniques, such as the well-known Forney algorithm.

[0042] The error corrector 20 receives the codeword c′(x) from the memory 22, along with each error location X_iand error value Y_ifrom the error location computation circuit 16 and the error values computation unit 18, respectively. It adds the error value Y_ito the corresponding erroneous symbol read from the error location X_ito produce on an output bus 34 an error-free symbol of codeword c(x).

[0043] All of the information needed to find errors and error values can be obtained from the syndrome values computed for Reed-Solomon and BCH codes. The D−1 syndrome values S[L], S[L+1], . . . , S[L+D−2] are assumed known for a minimum distance D code. This is the starting point of a full decoding process. The first step in the decoding process is to find the so-called error-locator polynomial σ(x) from the syndrome values, typically by applying a conventional Berlekamp-Massey algorithm or Euclidean algorithm.

[0044] Hereinafter, a conventional method of calculating the error locator polynomial according to the Euclidean algorithm will be briefly described. The Euclidean algorithm is a recursive algorithm by which a common divisor v(x) of two polynomials a(x) and b(x) is obtained and for producing an equation

[0045] s(x)a(x)+t(x)b(x)=v(x)  Eq. [4]

[0046] that expresses v(x) as a linear combination of a(x) and b(x). The equation may be rewritten as

[0047] t(x)b(x)=v(x) [mod a(x)]  Eq. [5]

[0048] The Euclidean algorithm solves the equation for t(x) and v(x) given the known a(x) and b(x) with a repetition of divisions of polynomials defined over a Galois field, and can be expressed as follows. That is, in the Euclidean algorithm, divisions of the two polynomials a(x) and b(x),

[0049] a(x)/b(x)=q₀(x) (quotient) . . . r_o(x) (remainder)

[0050] b(x)/r_o(x)=q₁(x) (quotient) . . . r₁(x) (remainder)

[0051] r_o(x)/r₁(x)=q₂(x) (quotient) . . . r₂(x) (remainder)

[0052] are repeated until predetermined conditions are satisfied. In application to BCH and RS codewords, the two polynomials are defined as a(x)=x^2tand b(x)=S(x), the syndrome polynomial. The term t(x) is defined as σ(x). The divisions are repeated until the degree of the remainder polynomial r(x) is reduced to less than t.

[0053] Referring to FIG. 2, a process of obtaining a solution (error locator polynomial σ(x)) 14 according to a conventional Euclidean algorithm is shown. Initially, in step 40, the process initializes a dividend value r(x)=x^2t, a divisor value w(x)=S(x)=syndrome values, v(x)=0 and σ(x)=1. The process determines if the degree of w(x)<t (step 42). If it is, the process is complete, that is, σ(x) is the error locator polynomial (step 44). If instead, at step 42, the degree of w(x) is found to be greater than t, then a quotient degree value d is computed as {degree of r(x)}−{degree of w(x)} (step 46). If d is greater than or equal to 0 (step 48), the process produces a quotient value p by dividing the leading term of r(x) by the leading term of w(x) (step 50). The process then computes r(x) as r(x)−px^dw(x) (step 52) and v(x) as v(x)−px^dσ(x) (step 54), and returns to step 46. If d is less than zero at step 48, the process interchanges r(x) with w(x) and v(x) with σ(x) (step 56), and returns to step 42.

[0054] Assume, for example, an ECC codeword with the capability to correct t=3 errors, x^2t=x⁶, and syndrome polynomial S(x)=α⁷x⁵+α⁰x⁴+α¹³x³+α¹¹x²+α¹⁴x+α⁷. Thus, referring back to FIG. 2, step 42, the degree of r(x) is equal to 6, which is not less than t=3. At step 46, d={degree of r(x)}−{degree of w(x)}=6−5=1. As d is greater than zero (step 48), the process computes p by dividing the leading term of r(x) by the leading term of w(x) as follows: ${\alpha }^7{\alpha }^0{\alpha }^{13}{\alpha }^{14}{\alpha }^{11}{\alpha }^7-\frac{\frac{{\alpha }^8}{\begin{array}{c}❘{\alpha }^{0}000000\\ {\alpha }^0{\alpha }^8{\alpha }^5{\alpha }^4{\alpha }^7{\alpha }^0\end{array}}}{{\alpha }^8{\alpha }^5{\alpha }^4{\alpha }^7{\alpha }^0}\mathrm{Deg}r\left(x\right)=4$

[0055] At step 50, p=α⁸and r(x) thus becomes remainder α⁰x⁶−α⁸x*w(x)=α⁸x⁵+α⁵x⁴+α⁴x³+α⁷x²⁺α⁰x at step 52 and v(x)=0−(α⁸x*1)=−α⁸x at step 54. Returning to step 46, d=5−5=0, so the division step 50 is again performed, as shown below. ${\alpha }^7{\alpha }^0{\alpha }^{13}{\alpha }^{11}{\alpha }^{14}{\alpha }^7-\begin{array}{cc}\frac{{\alpha }^8{\alpha }^1}{\frac{\begin{array}{c}❘{\alpha }^{0}000000\\ {\alpha }^0{\alpha }^8{\alpha }^5{\alpha }^4{\alpha }^7{\alpha }^0\end{array}}{{\alpha }^8{\alpha }^5{\alpha }^4{\alpha }^7{\alpha }^0}}& \mathrm{Deg}r\left(x\right)=4\\ -\frac{{\alpha }^8{\alpha }^1{\alpha }^{14}{\alpha }^{12}{\alpha }^0{\alpha }^8}{{\alpha }^{11}{\alpha }^9{\alpha }^{2}0{\alpha }^8}& \mathrm{Deg}r\left(x\right)=4\end{array}$

[0056] As shown above, p=α¹and r(x)=α¹¹x⁴+α⁹x³+α²x²+α⁰x+α⁸. The value of v(x)=−α⁸x−(α¹*1)=−α⁸x−α¹. Returning again to step 46, d=4−5=−1. Because d<0 (step 48), the process swaps r(x) with w(x), as well as v(x) and a(x), at step 56 and returns to step 42. Since the degree of w(x)=4 is not less than t (=3), the process performs another iteration of the computation of p, r(x) and v(x). The computation of p and r(x) is as follows: ${\alpha }^{11}{\alpha }^9{\alpha }^{2}0{\alpha }^8-\begin{array}{cc}\frac{{\alpha }^{11}{\alpha }^{14}}{\frac{\begin{array}{c}❘{\alpha }^7{\alpha }^0{\alpha }^{13}{\alpha }^{11}{\alpha }^{14}{\alpha }^7\\ {\alpha }^7{\alpha }^{15}{\alpha }^{13}0{\alpha }^4\end{array}}{{\alpha }^{10}{\alpha }^0{\alpha }^{11}{\alpha }^9{\alpha }^7}}& \mathrm{Deg}r\left(x\right)=4\\ -\frac{{\alpha }^{10}{\alpha }^8{\alpha }^{1}0{\alpha }^7}{{\alpha }^8{\alpha }^6{\alpha }^{9}0}& \mathrm{Deg}r\left(x\right)=3\end{array}$

[0057] The updated value of v(x)=1−(α¹¹x)*(−α⁸x−α¹)=α⁴x²+α²x+0. Since d=3−4=−1, the process swaps r(x) and w(x), as well as v(x) and σ(x). ${\alpha }^8{\alpha }^6{\alpha }^{9}0-\frac{\frac{{\alpha }^3}{\begin{array}{c}❘{\alpha }^{11}{\alpha }^9{\alpha }^{2}0{\alpha }^8\\ {\alpha }^{11}{\alpha }^9{\alpha }^{12}0\end{array}}}{0{\alpha }^{7}0{\alpha }^8}\mathrm{Deg}r\left(x\right)=2$

[0058] The value of σ(x)=σ₂(x)*α³x+σ₁(x)=α⁷x³+α⁵x²+α⁸x+α¹. As the degree of w(x) is now less than t (at step 42), the process is complete and the current value of σ(x) is the error locator polynomial.

[0059] It is understood that the degree of σ(x) corresponds to the number of errors. However, using the full Euclidean algorithm, and as can be seen from the example above, the degree is not known until the error locator polynomial itself has been calculated.

[0060] Unlike the conventional algorithm, the error locator polynomial degree determiner 26 (of FIG. 1) can accurately determine the number of errors in a codeword without determining the error-locator polynomial. As will appreciated from the description to follow, a modified algorithm performed by the determiner 26 saves time and storage space. That is, it does not need to carry out the coefficient update multiplications and does not need to save (i.e., store) and update the quotient values for the duration of the process. The quotient values may be discarded as soon as the quotient degrees are used to update the error locator polynomial degree, as they will not be needed to compute the error locator polynomial. Moreover, the dividend value can also be discarded as soon as the remainder value is computed.

[0061] Referring to FIG. 3, a Euclidean process of the error locator polynomial degree determiner 26 (of FIG. 1) that finds the error locator polynomial degree corresponding to the number of errors without computing the error locator polynomial coefficients is shown. The Euclidean process initializes r(x)=x^2t, a divisor w(x)=S(x)=syndrome values and an error locator polynomial degree value z=0 (step 60). The process determines if the degree of w(x)<t (step 62). If it is, the process is complete, indicating the degree z as the degree of the error locator polynomial (step 64). If instead, at step 62, the degree of w(x) is found to be greater than or equal to t, then a degree count indicator value y is set to 0 (step 66) and a quotient degree value d equal to the difference {degree of r(x))}−{degree of w(x)} is obtained (step 68). If d is greater than or equal to 0 (step 70), the process produces a quotient value p from the leading term of the dividend value r(x) and the leading term of divisor value w(x) by dividing the leading term of the dividend value r(x) by the leading term of divisor value w(x)(step 72). The process then generates r(x) from r(x)−px^dw(x) (step 74). If the process determines that y is equal to 0, then z becomes equal to z+d and y is set to 1 (step 76). The process then returns to step 68. If d is less than zero at step 70, then the process interchanges term r(x) with w(x) (step 78) and returns to step 62.

[0062] Other embodiments of the error locator polynomial degree determiner 26 of FIG. 1 are contemplated. For example, in one alternative embodiment of the error locator polynomial degree determiner 26, the number of errors is determined by evaluating a matrix (obtained from the syndromes). In this second approach, the computed syndromes S[L], S[L+1], . . . , S[L+D−2] are arranged to form the following specially shaped (trapezoid-shaped) “matrix”:

[0063] row0S[L], S[L+1], S[L+2], S[L+3], . . . , S[L+D−4], S[L+D−

[0064] 3], S[L+D−2]

[0065] row 1S[L+1], S[L+2], S[L+3], S[L+4], . . . , S[L+D−3], S[L+

[0066] D−2]

[0067] row 2S[L+2], S[L+3], S[L+4], S[L+5], . . . , S[L+D−2] . . .

[0068] row (t−1) S[L+t−1], S[L+t], S[L+t+1], . . . , S[L+D−2]  Matrix 1

[0069] In Matrix 1 above, t≦D/2 is the maximum number of errors to be determined. If, for example, t=5, D=10 and L=2, then Matrix 1 becomes

[0070] S[2], S[3], S[4], S[5], S[6], S[7], S[8], S[9],

[0071] S[10]

[0072] S[3], S[4], S[5], S[6], S[7], S[8], S[9], S[10]

[0073] S[4], S[5], S[6], S[7], S[8], S[9], S[10]

[0074] S[5], S[6], S[7], S[8], S[9], S[10]

[0075] S[6], S[7], S[8], S[9], S[10]  Matrix 2

[0076] For a t-row matrix, the number of errors that can be determined includes 0, 1, 2, . . . , t−1, as well as ≧t. The operations are row operations, which differ from row operations performed on a regular matrix (see, for example, pp. 32-37 of the Peterson and Weldon book referenced above) only insomuch that the rows of Matrix 2 are of different lengths. A row operation thus involves two rows of different lengths r[j] and r[k]. If r[j] is assumed greater than r[k] for j greater than k, the operation involves r[k] elements. Furthermore, the row with length r[j] is modified and XORed to the row with length r[k]. Given that there are t rows, row 0, row 1, . . . , row (t−1) with respective lengths D−1, D−2, . . . , D−t, the row reduction can be performed systematically.

[0077] Referring to FIG. 4, the matrix rank finding operation (or process) initializes a row number r_n=0 and an element number c_n=0 (step 80). The process first determines if row r_ncontains all zero values (step 82). If so, the process produces error number information indicating that there are r_nerrors (step 84). If row r_ndoes not contain all zero values at step 82, then the process determines if the c_n′th element in row r_nis equal to zero (step 86). If this element is zero, the process searches row r_n+1 to row t−1 to find a (preferably first) row with a nonzero element in position c_n(step 88). This first row is “row k”. If such a row k exists (step 90), the process interchanges row r_nwith row k (i.e., (row r_n, row k)=(row k, row r_n) (step 92) and returns to step 86. If row k does not exist, then the process increments c_nby 1 (step 94) and returns to step 88. If, at step 86, it is determined that the c_n′th element is nonzero, then the process uses row r_nto make the element c_nof rows k=r_n+1 to t−1 equal to zero (step 96) The process increments the value of r_nby 1 (i.e., r_nbecomes r_n+1) (step 98), and determines if r_nis equal to t (step 100). If r_nis equal to t, then the process produces error number information indicating that there are t or more errors (step 102). If r_nis not equal to t, then the process returns to step 82.

[0078] Referring to FIG. 5, the row operation 96 of FIG. 4 is shown in further detail. The process computes a constant K equal to (c_nof row k) divided by (c_nof row r_n) (step 104) and then multiplies row r_ntimes K (step 106). Once K*(row r_n) has been computed, the process XORs K*(row r_n) to row k (step 108). An example illustrating these row operation steps is provided below.

[0079] Here, mod 11 arithmetic is used (results are reduced by a multiple of 11) to make the final answer a value in the set of {0,1,2,3,4,5,6,7,8,9,10}, and values of rows r_nand k are

[0080] row r_n=0 0 7 4 9 8 1 6 5 and

[0081] row k=0 0 5 4 3.

[0082] Step 104:

[0083] K=5/7=5*(1/7)=5*([55+1]/7)=5*8=40=7 mod 11.

[0084] Step 106:

[0085] K*(row r_n)=7*(row r_n)=0 0 5 6 8 1 7 9 2

[0086] Step 108:

[0087] XOR of K*(row r_n) to row k=[7*(row r_n)]−row k=0 0 0 2 5.

[0088] The result of the row operation 96 (of FIGS. 4 and 5), illustrated in the above example, is that the c_n′th element is zero for row r_n+1, r_n+2, . . . , r_n+t−1. In other words, only row r_nhas a non-zero element at position c_n.

[0089] It will be appreciated that the operations of FIG. 5 are similar to those for finding the rank of a regular t*(D−1) matrix except that the rows are of unequal lengths. It will be further appreciated that other efficient rank finding algorithms can be modified (to take care of the unequal row lengths) to obtain the rank of the matrix.

[0090] For example, if Matrix 2 is reduced to $\begin{array}{cc}\begin{array}{ccccccccc}S\left[2\right],& S\left[3\right],& S\left[4\right],& S\left[5\right],& S\left[6\right],& S\left[7\right],& S\left[8\right],& S\left[9\right],& S\left[10\right]\\ 0,& X\left[4\right],& X\left[5\right],& X\left[6\right],& X\left[7\right],& X\left[8\right],& X\left[9\right],& X\left[10\right]& \\ 0,& 0,& 0,& 0,& 0,& 0,& 0,& 0& \\ 0,& 0,& 0,& 0,& 0,& 0,& 0& & \\ 0,& 0,& 0,& 0,& 0,& 0& & & \end{array}& \text{Matrix 3}\end{array}$

[0091] and it is assumed that x[j]'s are not all zeros, the rank of Matrix 3 is two. This rank implies that there are exactly two errors. If Matrix 2 is reduced to $\begin{array}{cc}\begin{array}{ccccccccc}S\left[2\right],& S\left[3\right],& S\left[4\right],& S\left[5\right],& S\left[6\right],& S\left[7\right],& S\left[8\right],& S\left[9\right],& S\left[10\right]\\ 0,& X\left[4\right],& X\left[5\right],& X\left[6\right],& X\left[7\right],& X\left[8\right],& X\left[9\right],& X\left[10\right]& \\ 0,& 0,& Y\left[6\right],& Y\left[7\right],& Y\left[8\right],& Y\left[9\right],& Y\left[10\right]& & \\ 0,& 0,& 0,& 0,& 0,& 0& & & \\ 0,& 0,& 0,& 0,& 0& & & & \end{array}& \text{Matrix 4}\end{array}$

[0092] with X[j]'s not all zero and Y[k]'s not all zeros, the rank is three, thus implying a rank of 3.

[0093] If it is desirable to produce error number information that indicates whether or not the number of errors is below a given threshold h, where h is less than or equal to t, the process needs to use an h row matrix. After the row reductions have been completed, only a last row, row h−1, needs to be examined: if it is an all zero row, then there are fewer than h errors; otherwise, there are h or more errors. This type of threshold detection is particularly useful for a threshold of h=t, as t is the maximum number of errors which can be corrected. Thus, for a threshold of h=t, for example, the process 26 shown in FIG. 4 can be modified for this threshold detection by determining if r_nis equal to t−1 at step 100, returning to step 86 for further row reduction operations if r_nis not equal to t−1 and returning to step 82 to examine the row as a potential all zero row if r_nis equal to t−1. If the row has all zero values, the process indicates less than t errors (instead of r_nerrors) at step 84. If the row has nonzero values, the process indicates t or more errors (i.e., step 102).

[0094] That the number of errors may be determined accurately using the above-described matrix method is supporting by the following. It is well-known that the number of errors of an error pattern, t, corresponds to the degree of the error locator polynomial and that for t<D/2, the error locator polynomial σ(x) is unique. Furthermore, if σ(t)*S[L+j]+σ(t−1)*S[L+j+1]+σ(t−2)*S[L+j+2]+ . . . +σ(2)*S[L+j+t−2]+σ(1)*S[L+j+t−1]+σ(0)*S[L+j+t]=0 for j=0, 1, 2, . . . , D−1−t, then it follows that $\begin{array}{cc}\left[\sigma \left(t\right)\sigma \left(t-1\right)\dots \sigma \left(1\right)\sigma \left(0\right)\right][\begin{array}{cccc}S\left[L+j\right]& S\left[L+j+1\right]& \dots & S\left[L+j+t\right]\\ S\left[L+j+1\right]& S\left[L+j+2\right]& \dots & S\left[L+j+t+1\right]\\ \dots & \dots & \dots & \\ S\left[L+j+t\right]& S\left[L+j+t+1\right]& \dots & S\left[L+j+2t\right]\end{array}]=0.& \text{Eq. [6]}\end{array}$

[0095] The matrix of Eq. [4] is a sub-matrix of Matrix 1. The rank of the matrix of Eq. [4] is t. Therefore, the row operations of Matrix 1 result in t non-zero rows. One way to ensure that the error locator polynomial is of degree t is to perform rank computation on every possible sub-matrix of Matrix 1 of (t+1)*(t+1) to determine that each of the ranks is indeed t. This rank computation, which is time-consuming, is equivalent to the more efficient matrix evaluation process 26 of FIG. 4 as earlier presented.

[0096] Referring to FIG. 6, a data storage system 110 for employing the determiner 26 (from FIG. 1) to generate an error number is shown. The data storage system 110 includes a read/write controller 112 for reading codewords from and writing codewords to a storage device 114, which includes a storage medium such as a magnetic tape. In addition to the read/write controller 112 and the storage device 114, the data storage system 110 includes the decoder 10 (from FIG. 1) and a threshold comparison circuit 116. Although depicted as a circuit, the threshold comparison may be implemented in software as a threshold comparison unit 116. The threshold comparison circuit (or unit) 116 generates a pass/fail signal to the read/write controller 112 in response to the number of errors received from the decoder 10 and a predetermined error threshold supplied by the controller 28. Alternatively, the predetermined error threshold may be stored in the threshold comparison circuit. Although the decoder 10 is identical to the decoder 10 of FIG. 1, only the error count unit 31 (with its internal functional units, the syndrome generator 12, the error locator polynomial degree determiner 26 and the controller 28) is shown for purposes of simplification.

[0097] Still referring to FIG. 6, the read/write controller reads a codeword from a storage location on the storage device 114 and transfers the codeword to the syndrome generator 12 of the error count unit 31. The syndrome generator 12 process the codeword to produce a syndrome polynomial S(x), which is associated with an error locator polynomial. The error locator polynomial determiner 26 then determines from the syndrome polynomial the degree of the error locator polynomial without computing the error locator polynomial coefficients, preferably using one of the techniques described with reference to FIGS. 3 and 4. The determiner 26 identifies as an output value 117 the degree as the number of errors in the codeword and sends the error number output value, along with appropriate control signals 118 (which include the predetermined error threshold) from the controller 28 to the threshold comparison circuit 116. In response to these inputs, the threshold comparison circuit 116 compares the number of errors to the predetermined error threshold and provides to the read/write controller 112 a “fail” indication 119 if the threshold comparison circuit 116 determines that the number exceeds the predetermined error threshold. The read/write controller can then take appropriate action based on this failure indication. For instance, if the storage medium is magnetic tape, which is inherently less reliable than, say, magnetic disk, the read/write controller can immediately re-write the codeword c(x) to the storage location from which the codeword was read, or to a different storage location on the tape.

[0098] It will be appreciated that the computational process of determiner 26 may be implemented in hardware, or in software executing on a processing unit (such as a microprocessor). It will be further appreciated that the error count unit 31 of the present invention need not reside in a decoder as shown in FIG. 1. It may be employed in hardware and/or software which does not routinely perform error decoding/correction operations. Such an application is a storage device test unit, an example of which is shown in FIG. 7.

[0099] Referring to FIG. 7, a test unit 120 includes a read/write controller 122 for writing codewords c(x) to and reading codewords c(x)+e(x) from a storage device 124 (shown in dashed lines), such as a disk, under test. Each retrieved codeword is provided to the syndrome generator 12, which is controlled by the controller 28 to send the syndromes S(x) to the error locator polynomial degree determiner 26, which operates to in the manner described above (with reference to FIG. 3 or FIG. 4). The determiner 26 is directed by the controller 28 to provide the error number to a threshold comparison circuit (or unit) 126 to produce a pass/fail determination based on whether the error number exceeds a predetermined error threshold. If the error number is above threshold, the threshold comparison circuit 126 provides a “fail” indication to the read/write controller. The read/write controller then determines, based on the pass/fail determinations for each of the read codewords (and thus their associated storage locations), if the disk as a whole is bad. That is, is determines if a predetermined number of one or more of the read codewords has an above-threshold number of errors. For such a determination, the read/write controller communicates the disk failure to an operator display device (not shown), which indicates to the operator the disk's rejected status.

[0100] It will be understood that the overall disk test methodology, such as the number of read storage locations, e.g., sectors, of the read/write controller is a matter of design choice and may be implemented using known techniques. For example, one option may be to test each sector on the disk and determine the disk's failure status based on a predetermined number of one or more failed sectors. Another approach may involve examining a number of one or more sectors per track for pass/fail status and determining the disk's status from a number of one or more bad tracks per disk. For the latter technique, the predetermined error threshold would be set to a lower level for a tighter test limit to compensate for the fewer number of sectors being tested.

[0101] Alternatively, with respect to both FIGS. 6 and 7, the determiner 26 could produce a threshold detection outcome (e.g., less than t) in the manner described with reference to FIG. 5.

[0102] It is to be understood that while the invention has been described in conjunction with the detailed description thereof, the foregoing description is intended to illustrate and not limit the scope of the invention, which is defined by the scope of the appended claims. Other embodiments are within the scope of the following claims.