Fast and efficient decoder for low-distance RS and BCH codes

By designing the circuit of the syndrome calculator, position finder and error corrector, the problem of low error correction efficiency in the prior art RS and BCH code decoders in high bandwidth communication is solved, and efficient error correction and fast Error positioning.

CN119010924APending Publication Date: 2024-11-22CREDO TECHNOLOGY GROUP LTD
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202410274709.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2023-05-17
Filing Date
2024-03-11
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

Existing RS and BCH code decoders are difficult to implement parallelization or iterative programs at high data rates, resulting in low error correction efficiency in high bandwidth communication.

Method used

A circuit and method including a syndrome calculator, a position finder and an error corrector are designed to calculate the syndrome value by obtaining the product of the received message vector and the parity check matrix, and use a logic gate array and a lookup table to determine the error. Position and polynomial roots to achieve error correction.

Benefits of technology

It improves the efficiency and speed of error correction, reduces delay, and is suitable for error correction of BCH and RS codes in high-bandwidth communication, enhancing the performance of the decoder.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119010924A_ABST
    Figure CN119010924A_ABST
Patent Text Reader

Abstract

The invention discloses a fast and efficient decoder for low distance RS and BCH codes. An illustrative decoder includes a syndrome calculator, a position finder, and an error corrector. The syndrome calculator has a logic gate array for obtaining a syndrome value as a product of the received message vector and the parity check matrix, the syndrome value comprising at least three ten-bit syndrome values S1, S2 and S3. A position finder derives an error number from the syndrome value and uses a second logic gate array for obtaining two polynomial roots corresponding to a product of the syndrome value vector and a quadratic solution matrix when the error number is 2, the quadratic solution matrix corresponding to a determination of a trailing coefficient value s of a quadratic equation, a determination of a root of the quadratic equation, and inversion of the variable substitution. The location finder further determines a bit index for each polynomial root.
Need to check novelty before this filing date? Find Prior Art

Description

Background Art

[0001] Reed-Solomon (RS) and Bose-Chaudhuri-Hocquenghem (BCH) codes are commonly used for forward error correction (FEC) in modern communication channels, introducing enough redundancy to enable the receiver to correct noise-induced symbol detection errors. RS and BCH codes treat each data block as a set of polynomial coefficients. This message polynomial is multiplied by a "generator" polynomial known to both the encoder and the encoder to determine the "codeword" polynomial corresponding to the message to be sent. The generator polynomial is derived based on the expected length of the codeword and the expected Hamming distance between codewords.

[0002] Note that RS or BCH codes (and decoders) are often only one component of a complex FEC strategy. For example, the International Telecommunication Union standard ITU-T G.709.2 / Y.1331.2 (July 2018) specifies a FEC strategy for the OTU4 long-haul interface (as part of the Optical Internetworking Forum Implementation Agreement OIF-400ZR-01.0) that includes an error demodulator function, staircase FEC codes, and a scrambler. OIF-400ZR-01.0 further enhances the ITU-T standard by adding a convolutional interleaver and an internal Hamming code. In both cases, the staircase code contains the BCH code as a component, so one or more BCH decoders are required at the receiver.

[0003] There are RS and BCH code variants, such as shortening, puncturing or extending the code message to reduce the word length, remove redundancy or add redundancy respectively, while still allowing the reuse of existing decoder designs. The ITU-T standard specifies the use of such shortened extended BCH codes. This code takes a block of k = 990 message bits and multiplies it with a generator or parity check matrix to obtain 32 parity check bits, producing n = 1022 codeword bits with a minimum Hamming distance of d = 8. To summarize these parameters, the code can be called a BCH (1022, 990, 8) code.

[0004] There are many RS and BCH decoding techniques, most of which are: first derive the error syndrome value S from the received version of the codeword polynomial i , determine the number of symbol errors, use the error syndrome value to determine the coefficients of the appropriate error locator polynomial, operate on the error locator polynomial to find its roots (indicating the error location), calculate the error value as needed, and then correct the error. Existing decoder techniques use iterative procedures that cannot be parallelized or implemented at higher data rates. Summary of the invention

[0005] Therefore, circuits and methods for correcting bit errors in a received version of an RS or BCH encoded bit stream are disclosed herein. An illustrative circuit includes: a syndrome calculator, a position finder, and an error corrector. The syndrome calculator has a first logic gate array for obtaining a syndrome value as a product of a received message vector and a parity check matrix, the syndrome value including at least a first ten-bit syndrome value S1, a twentieth-bit element syndrome value S2, and a thirtieth-bit syndrome value S3. The position finder derives an error number from the syndrome value and includes a second logic gate array for obtaining two polynomial roots corresponding to the product of the syndrome value vector and a quadratic solution matrix when the error number is 2, the quadratic solution matrix corresponding to a determination result of a trailing coefficient value s of a quadratic equation, a determination result of a root of the quadratic equation, and an inversion of a variable substitution. The position finder further includes an index circuit for determining a bit index of each polynomial root. The error corrector receives, for each received message vector, a set of zero or more bit indices representing error positions in the received message vector.

[0006] An illustrative error correction method includes: obtaining a syndrome value corresponding to the product of a received message vector and a parity check matrix, the syndrome value including at least a first ten-bit syndrome value S1, a twentieth-bit syndrome value S2, and a thirtieth-bit syndrome value S3; converting the syndrome value to a set of zero or more polynomial roots representing the position of an error in the received message vector; and determining the bit index of each polynomial root in the set. The conversion operation includes: deriving an error number from the syndrome value; when the error number is 2, using a logic gate array to obtain two polynomial roots corresponding to the product of the syndrome value vector and a quadratic solution matrix, the quadratic solution matrix corresponding to the trailing coefficient value s of the quadratic equation, determining the root of the quadratic equation, and inverting the variable substitution.

[0007] Another illustrative error correction method includes: obtaining a syndrome value corresponding to the product of a received message vector and a parity check matrix, the syndrome value including at least a first ten-bit syndrome value S1, a twentieth-bit syndrome value S2, and a thirtieth-bit syndrome value S3; converting the syndrome value to a set of zero or more polynomial roots corresponding to error positions in the received message vector; and determining a bit index of each polynomial root in the set. The conversion includes: deriving an error number from the syndrome value; when the error number is three, using a first lookup table with supporting logic gates to obtain three polynomial roots, when The three polynomial roots correspond to the cubic equation x 3 + the root of d, where d can be expressed as The depth of the first lookup table does not exceed 341.

[0008] Each of the foregoing embodiments may be implemented individually or in combination, and may be implemented using one or more of the following features in any suitable combination: 1. The inversion of variable substitution may be expressed as x=S1·t, and the trailing coefficient value may be expressed as 2. The first logic gate array and the second logic gate array are respectively composed of a set of logic AND gates and a set of logic XOR gates, the set of logic AND gates is used to implement bit-by-bit multiplication, and the set of logic XOR gates is used to implement bit-by-bit addition. 3. The syndrome value includes an error parity value P. 4. The error corrector is configured to accumulate parity increments of a parity check matrix based on a bit index, and is configured to calculate an inverse of a received message vector bit corresponding to the bit index if the parity increment indicates that the bit error is correctable. 5. The position finder further includes a first lookup table having supporting logic gates for accumulating a parity increment ... parity check matrix based on a bit index, and is configured to calculate an inverse of a received message vector bit corresponding to the bit index if the parity increment indicates that the bit error is correctable. When we get three polynomial roots, the three polynomial roots correspond to the cubic equation x 3 + the root of d, where d can be expressed as and the depth of the first lookup table does not exceed 341. 6. The first lookup table contains a single polynomial root or error position, and the supporting logic gates derive the remaining polynomial roots or error positions from the single polynomial root or error position. 7. The position finder further includes a second lookup table with supporting logic gates for deriving the remaining polynomial roots or error positions when the number of errors is three and When we get three polynomial roots, the three polynomial roots correspond to the cubic equation x 3 +x+d, where d can be expressed as and the depth of the first lookup table does not exceed 170. 8. The second lookup table contains two polynomial roots or error positions, and the support logic gates of the second lookup table derive the remaining polynomial root or error position from the two polynomial roots or error positions contained in the second lookup table. 9. The position finder is configured to provide the syndrome value S1 as the polynomial root when the error number is 1. 10. The received message vector contains exactly 1022 bits, and the parity check matrix H comp It has 32×1022 binary elements, representing the composite field GGF((2 5 ) 2 ) is the power of the root of the original polynomial in . 11. Derivation of the error number includes: if the syndrome values ​​are all zero, then the error number is determined to be zero; if the syndrome value S1 is non-zero and the remaining syndrome values ​​are all zero, then the error number is determined to be 1; if Then the error number is determined to be 2; otherwise, the error number is determined to be 3. 12. The method includes verifying whether the quadratic equation has a root by determining that the number of bits of the binary representation of the trailing coefficient s is s7=0 if the error number is 2. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1is a perspective view of an illustrative active Ethernet cable.

[0010] Figure 2 is a functional block diagram of an illustrative AEC.

[0011] Figure 3 is an architectural diagram of a communication link including illustrative cables.

[0012] Figure 4 is a block diagram of an illustrative transmit chain and its corresponding receive chain.

[0013] Figure 5 is a block diagram of an illustrative staircase code decoder.

[0014] Figure 6 is a block diagram of an illustrative BCH decoder.

[0015] Figure 7 Shows the relationship between underlying fields and composite fields.

[0016] Figure 8 is a flow chart of an illustrative BCH decoding method. DETAILED DESCRIPTION

[0017] Although specific embodiments are presented in the drawings and the following description, it should be kept in mind that they do not limit the present disclosure. Instead, they provide a basis for one of ordinary skill in the art to identify alternatives, equivalents, and modifications that are included within the scope of the appended claims.

[0018] To provide an illustrative context for understanding the myriad applications of the disclosed decoder, Figure 1 1 is a perspective view of an illustrative 400 Gbps cable that can be used to provide high bandwidth communication links between devices in a routing network, such as a routing network for data centers, server farms, and interconnect exchanges. The routing network can be part of or can include, for example, the Internet, a wide area network, or a local area network. The linked devices can be computers, switches, routers, etc. The cable includes a first connector 100 and a second connector 101 connected via one or more optical fibers in an electrical cord 106. The one or more optical fibers can be configured for unidirectional or bidirectional communication of wavelength division multiplexed optical signals, respectively.

[0019] To achieve robust performance over extended cable lengths, each connector 100, 101 may include a powered transceiver that performs electrical-to-optical signal conversion combined with clock and data recovery (CDR) and remodulation of data streams in each direction. Powered transceivers are also referred to as data recovery and remodulation (DRR) devices. In at least one contemplated embodiment, the cable connectors 100, 101 are quad small form factor pluggable double density (QSFP-DD) transceiver modules, and the transceiver modules exchange 400GAUI-8 data streams with a host.

[0020] Figure 2 yes Figure 1 The connector 100 includes a plug 200 adapted to fit a first host device 302 (see Figure 3 ) in a standard-compatible Ethernet port in the connector 100 to receive an electrical input signal carrying an outbound data stream from the host device and provide an electrical output signal carrying an inbound data stream to the host device. Similarly, the connector 101 includes a plug 201 adapted to the Ethernet port of the second host device 304. The connector 100 includes a first DRR device 202 for performing CDR and remodulation on data streams entering and leaving the cable at the connector 100, and the connector 101 includes a second DRR device 204 for performing CDR and remodulation on data streams entering and leaving the cable at the connector 101. The fiber optic transceiver chips 240, 242 provide conversion between electrical signals and optical signals to transmit data streams on the optical fiber 107 coupled between the connectors 100, 101. The DRR devices 202, 204 can be integrated circuits mounted on a printed circuit board and connected to edge connector contacts and optical transceivers 240, 242 via circuit board traces. Optical couplers connected to the transceivers 240 , 242 may connect the transceivers to the optical fiber 107 .

[0021] In at least some contemplated embodiments, the printed circuit boards also each support a microcontroller unit (MCU) 206. Each DRR device 202, 204 is coupled via a first two-wire bus to a corresponding MCU device 206 that configures the operation of the DRR device. At power-on, the MCU device 206 loads equalization parameters and / or other operating parameters from a flash memory 207 into a configuration register 208 of the DRR device. A host device can access the MCU device 206 via a second two-wire bus that operates according to an I2C bus protocol and / or a faster MDIO protocol. With this access to the MCU device 206, the host device can adjust the operating parameters of the cable and monitor the performance of the cable.

[0022] Each DRR device 202, 204 includes a transmitter and receiver set 220 for communicating with a host device and a transmitter and receiver set 222 for communicating via an optical transceiver and an intermediate optical fiber. The illustrated host-oriented transceiver 220 supports eight channels (400GAUI-8) for bidirectional communication with a host device. In other contemplated embodiments, the host-oriented transceiver 220 supports other data rates and channel configurations. The DRR device includes a memory 224 for providing a first-in-first-out (FIFO) buffer between the transmitter and receiver sets 220, 222. The embedded controller 228 coordinates the operation of the transmitter and receiver by, for example, setting initial equalization parameters and ensuring that the training phase is completed across all channels and links before enabling the transmitter and receiver to enter the data transmission phase. The embedded controller 228 uses a register set 208 to receive commands and parameter values ​​and provide responses that potentially include status information and performance data.

[0023] Figure 2 The illustrative cable may be part of a point-to-point communication link between two host devices 302, 304, such as Figure 3 The architecture diagram is shown in the figure. Figure 3 The use of the ISO / IEC model for open systems interconnection (see ISO / IEC 7498-1: 1994.1) is shown for communication via channel 306 over a physical medium such as optical fiber. The interconnection reference model employs a hierarchy of layers with defined functions and interfaces to facilitate the design and implementation of compatible systems by different teams or vendors. Although not a requirement, it is expected that the higher layers in the hierarchy will be implemented primarily by software or firmware operating on a programmable processor, while the lower layers will be implemented using microcode and / or application-specific hardware.

[0024] The application layer 308 is the top layer in the model and represents user applications or other software that operate different systems that require facilities for transmitting messages or data. The presentation layer 310 provides such applications with a set of application programming interfaces (APIs) that provide formal syntax as well as services for data conversion (e.g., compression), establishing communication sessions, connectionless communication modes, and negotiation to enable application software to identify available service options and select from them. The session layer 312 provides services for coordinating data exchanges, including: session synchronization, token management, full-duplex or half-duplex mode implementation, and establishing, managing, and releasing session connections. In connectionless mode, the session layer may only map between session addresses and transport addresses.

[0025] Transport layer 314 provides services for the following: multiplexing, end-to-end sequence control, error detection, segmentation, block, cascading, flow control on each connection (including suspension / recovery) and implementation of end-to-end quality of service specifications. The focus of transport layer 314 is end-to-end performance / behavior. Network layer 316 provides routing services, determines the link for end-to-end connection, and acts as a relay service to couple such links together when necessary. Data link layer 318 acts as an interface to physical connection, thereby providing delimitation, synchronization, sequence and flow control across physical connection. It can also detect and optionally correct errors that occur across physical connection. Physical layer 322 provides mechanical, electrical, functional and program means to activate, maintain and deactivate channel 306, and uses channel 306 for bit transmission across physical medium.

[0026] The data link layer 318 and the physical layer 322 are slightly subdivided and modified by IEEE Standard 802.3-2015, which provides a medium access control (MAC) sublayer 320 in the data link layer 318 to define the interface with the physical layer 322, including the frame structure and transfer syntax. Within the physical layer 322, the standard provides various possible subdivisions, such as Figure 3 One subdivision is shown, which includes an optional harmonization sublayer 324, a physical coding sublayer (PCS) 326, a forward error correction (FEC) sublayer 328, a physical medium attachment (PMA) sublayer 330, and a physical medium dependent (PMD) sublayer 322.

[0027] Optional reconciliation sublayer 324 is mapped only between interfaces defined for MAC sublayer 320 and PCS sublayer 326. PCS sublayer 326 provides alignment mark insertion / removal, FEC, and framing with synchronization and training sequences. PMA sublayer 330 provides symbol encoding / decoding, filtering, conversion between analog signal format and digital signal format. PMD sublayer 332B specifies optical transceiver conversion between transmission / reception channel signals and corresponding electrical signals. Socket 336 is also shown as a part of PMD sublayer 332 to represent physical network interface port.

[0028] Connectors 100, 101 have plugs 200, 201 representing edge connectors that mate with receptacles 336 of two host devices 302, 304. In each connector, the DRR device can implement a host-facing physical layer 322A, a center-facing physical layer 322B, and a data link layer 340 that bridges the two physical layers together. In some embodiments, one or more internal sublayers (e.g., PCS, reconciliation, MAC) in each connector are completely bypassed or omitted to reduce area requirements and / or reduce power. In many cases, more information about the operation of the sublayers, as well as the electrical and physical specifications of the connection between the node and the communication medium (e.g., pin layout, line impedance, signal voltage and timing), and the physical specifications for the communication medium itself (e.g., attenuation, dispersion limit) can be found in current standards, and any such details should be considered to be within the knowledge of ordinary technicians in the field.

[0029] Figure 4 A more detailed block diagram of an illustrative transmit and receive chain in the PCS sublayer is provided. Figure 4 The PCS sublayer transmission chain in receives the 400GMIIPCS data stream. According to the standard, the PCS data stream has been encoded with a transmission code that provides DC balance and enables timing recovery. The transcoder module 402 modifies the transmission code from 64b / 66b code to 256b / 257b code that is more suitable for use with FEC strategies. The receiving chain includes a reverse transcoder module 403 to reverse this operation.

[0030] The Generic Mapping Program (GMP) modules 404, 405 provide translation between the local system clock domain and the line clock domain, respectively, and typically provide word stuffing to accommodate clock rate mismatches. The Alignment Marker Insertion module 406 provides alignment marks that can be aligned between different lanes of the data stream. The Detector module 407 detects and removes alignment marks. Note that these modules can cooperate with the Alignment Marker Insertion / Removal modules in other sublayers to preserve the alignment mark content throughout the transcoding process.

[0031] The cyclic redundancy check calculation module 408 adds checksum information to the data stream, which is verified by the check module 409 at the receiving end to detect data corruption. The error de-correlation interleaving module 410 reallocates the symbols of the data stream, and this operation is reversed by the de-interleaving module 411 at the receiving end. The staircase encoder module 412 implements the staircase code with BCH component codes. The staircase decoder module 413 reverses this operation, which is described in detail below.

[0032] Module 414 appends padding bits to align the staircase codewords with the 400ZR frame boundaries of 400ZR and applies a predefined scrambling mask. Module 415 reverses these operations at the receiving end. Convolutional interleaver 416 spreads bits out of the codewords to increase resilience to burst errors. Deinterleaving module 417 reverses this operation at the receiving end.

[0033] Hamming encoder module 418 applies double extended Hamming code SD-FEC (128, 119) to provide additional net coding gain. Module 419 performs decoding operations at the receiving end. Symbol mapper module 420 maps the encoded data stream 8 bits at a time into dual polarization 16-point quadrature amplitude modulation (DP-16QAM) symbols, which are then interleaved by symbol interleaver module 422. The receive chain includes corresponding modules 421, 423 to reverse these operations.

[0034] To aid in timing synchronization and clock recovery, module 424 provides a frame alignment word (FAW) at the beginning of each superframe, which includes 49 subframes, each with a training sequence provided by module 424. Module 424 further inserts pilot symbols every 32 symbols. In the receive chain, module 425 uses the FAW to detect alignment of the superframe and then removes the pilot symbols, training sequence, and FAW.

[0035] Modules 426, 427 represent the PMA sublayer, where digital signal processing can be performed for spectrum shaping and signal equalization and conversion between digital and analog signal domains. At the receiving end, module 427 performs clock recovery as part of analog-to-digital conversion. Modules 428, 429 represent the PMD sublayer, where the optical transceiver converts the electrical transmit signal into an optical signal in the optical fiber and converts the received optical signal into an electrical receive signal at the receiving end.

[0036] The operation of the decoder module 413 is as follows: Figure 5 The memory 502 receives the encoded input stream, which is conceptually organized into blocks B in the decoding window 504. i Each block has data bits on the left and parity bits of codewords 508, 510 on the right, which span the data bits and the previous data block. A set of parallel BCH decoders 506 operates alternately on the vertical codewords 508 and horizontal codewords 510, iterating forward and backward through the blocks in the decoding window 504. In theory, the iteration can be repeated until there are no further error corrections. In practice, a predetermined number of iterations are performed before the decoding window 504 is moved.

[0037] The memory 502 can be organized as bytes or words, resulting in memory locations being accessed in an easily readable manner during initial processing of a block, but in a more decentralized manner when the block is subsequently processed. To facilitate subsequent processing, the decoder 506 can form a transpose of each block during initial processing, storing the transposed blocks in a set of transposed memory blocks 512. Further implementation details can be found in the literature, including, for example, D. Truhachev et al., "Efficient Implementation of 400 Gbps Optical Communications FEC", IEEE Transmission Circuits and Systems-I, V68n1, January 2021.

[0038] It is desirable that each BCH decoder 506 be implemented as efficiently as possible to achieve high throughput with minimal power consumption.

[0039] Figure 6 is a schematic diagram of an illustrative BCH decoder implemented conceptually in three parts. Syndrome calculator 602 operates on the received message R to obtain syndrome values ​​S. From these syndrome values, position finder 604 determines the error number and position i for each error (or determines that the error number exceeds the correction capability of the code). Using the error positions, error corrector 606 modifies the received message R to produce a corrected message .

[0040] Conventional implementations of position finder 604 employ an iterative procedure to compute and decompose the error position polynomial, such as the Berlekamp-Massey algorithm using a Chien search. Such iterations can be difficult to accommodate when decoding high bandwidth data streams. The enhanced decoding method disclosed below may provide at least four potential advantages over existing decoder implementations: (1) fast, non-LUT based determination of whether the error polynomial has any roots is performed using properties of the tracking function; (2) the error polynomial decomposition operation is simplified by using a composite field; (3) the basis transforms for the composite field are precomputed, so no overhead is incurred; and (4) the LUT used to find the cubic roots is much smaller.

[0041] In the description Figure 8 Enhanced decoding method for harmonic coders Before, we explain the BCH codes provided in the standard and the algebraic techniques for deriving enhanced decoding methods. These techniques are applicable to all BCH and RS codes and are particularly suitable for such codes with low correction power, for example, d < 4 or t < 9.

[0042] The underlying field of the (1022, 990) extended BCH code used in the ITU-T G.709.2 / Y.1331.2 standard is GF(2 10 ), which can be expressed by the primitive polynomial p(x)=x 10 +x 3+1 construction. α is set to the root of the original polynomial p(x). GF(2 10 ) can be represented as α i , 0≤i≤1022, which we call "power representation", where α 1023 =α 0 = 1. A consequence of the field construction is that its elements can be expressed as weighted sums of the first ten powers, i.e., we can write α i =b9α 9 +b a a 8 +…+b0, 0≤i≤1022, where the coefficients are binary. For ease of indexing, we will use the integer l=b92 9 +b82 8 +…+b0 is called α i The "binary" representation of . We define log(l) = i.

[0043] BCH encoding is done using a generator matrix defined from these field elements. To implement systematic encodings where a codeword consists of the original data word concatenated with a set of parity check symbols, the generator matrix columns can be permuted so that a portion of the generator matrix resembles the identity matrix. The standard defines a permutation function Π on integers i, 0≤i≤509 d To achieve this result. In the following, ∏ d (M:M+N)=K:K+N is ∏ d (M) = K, ∏ d (M+1)=K+1,…,∏ d Abbreviation for (M+N)=K+N. d The values ​​of are specified in Table 1 below. Table 1 ∏ d Value <![CDATA[∏ d (0:7)=478:485]]> <![CDATA[∏ d (8)=0]]> <![CDATA[∏ d (9:11)=486:488]]> <![CDATA[∏ d (12)=1]]> <![CDATA[∏ d (13)=489]]> <![CDATA[∏ d (14:16)=2:4]]> <![CDATA[∏ d (17:19)=490:492]]> <![CDATA[∏ d (20)=5]]> <![CDATA[∏ d (21)=493]]> <![CDATA[∏ d (22:24)=6:8]]> <![CDATA[∏ d (25)=494]]> <![CDATA[∏ d (26:32)=9:15 <!-- 6 -->]]> <![CDATA[∏ d (33:35)=495:497]]> <![CDATA[∏ d (36)=16]]> <![CDATA[∏ d (37)=498]]> <![CDATA[∏ d (38:40)=17:19]]> <![CDATA[∏ d (41)=499]]> <![CDATA[∏ d (42:48)=20:26]]> <![CDATA[∏ d (49)=500]]> <![CDATA[∏ d (50:64)=27:41]]> <![CDATA[∏ d (65:67)=501:503]]> <![CDATA[∏ d (68)=42]]> <![CDATA[∏ d (69)=504]]> <![CDATA[∏ d (70:72)=43:45]]> <![CDATA[∏ d (73)=505]]> <![CDATA[∏ d (74:80)=46:52]]> <![CDATA[∏ d (81)=506]]> <![CDATA[∏ d (82:128)=53:99]]> <![CDATA[∏ d (129)=507]]> <![CDATA[∏ d (130)=100]]> <![CDATA[∏ d (131)=508]]> <![CDATA[∏ d (132:256)=101:225]]> <![CDATA[∏ d (257)=509]]> <![CDATA[∏ d (258:509)=226:477]]>

[0044] Consider a function f(i) that maps an integer i, 1≤i≤1023, to a column vector, where where β i =α log(i) ,and For the binary representation of l, β i ,and is the complement of x. Then the generator matrix can be obtained as follows. First, the column vector is similar to: H=[f(1021)f(1022)f(1)…f(510)f(511+∏ -1 (0))…f(511+∏ -1 (509)] Then replace each element in the first three rows with its corresponding 10-bit binary representation and perform the basic row operations of H (over GF(2)) to obtain its row reduced echelon form H with the identity matrix on the right ENC =[P F ; I]. The generator matrix is ​​then G = [I; P], and the resulting 990x32 matrix P provides the parity check generator matrix for the encoder.

[0045] The encoder can be implemented as a 1x990 vector (990-bit message) multiplied by a 990×1022 matrix (generator matrix). The multiplication and addition of matrix elements are performed over GF(2), where AND gates and XOR gates correspond to multipliers and adders, respectively. Therefore, the encoder can be implemented as an array of AND-XOR gates.

[0046] GF(2 10 ) via p(x)=x 10 +x 3 The multiplicative inverse operation of the polynomial basis constructed by +1 is complex. If implemented via a lookup table (LUT), the depth of the LUT will be 1023, which may deteriorate the critical path. However, GF(2 10 ) and the composite field GF((2 5 ) 2 ) isomorphic, allowing us to exploit composite fields to simplify multiplicative inverse operations. In addition, composite fields simplify solving quadratic equations because the composite field GF((2 5 ) 2 ) can be decomposed into subfields GF(2 5 ) in several operations.

[0047] To use composite fields in the ITU-T standard (1022, 990) extended BCH encoder / decoder, we must consider basis conversion. We note that the generator polynomial of the (1022, 990) BCH code is GF(2 10 ) via p(x)=x 10 +x 3 +1The polynomial basis and composite fields of the construction can be the same, because α, α 3 , α 5 The minimal polynomial of can be the same, that is, it does not depend on the choice of the underlying field.

[0048] Therefore, the (1022, 990) BCH code with composite field representation is the same as the code with polynomial representation defined in the ITU-T standard. However, for the decoder, we do not directly use the matrix H defined in the standard. Instead, we convert the matrix H into H comComposite fields. Transformations can be performed as follows: Transformations can be performed as follows: .H com The last two rows of are the same as those in H. Each column of .H excluding the last two rows can be considered as GF(2 10 ) in the polynomial basis representation, namely and Will and Convert to H com A composite field representation of .

[0049] H vs H com The conversion between is pre-computed, i.e., no additional hardware cost is required. The decoding, including syndrome calculation, error location polynomial discovery, and error location polynomial decomposition, can be performed on the composite field GF((2 5 ) 2 ). Next, we will introduce composite field construction.

[0050] Let us first review the composite field GF(2 mn ) algebraic structure. GF(2 mn ) can be obtained by n ) is constructed by irreducible polynomials of degree m on GF(2 n ) is called the base field, and the base field can be obtained by n ) is constructed by using irreducible polynomials of degree n on GF(2). When gcd(m,n)=1, we can use two irreducible polynomials on GF(2) to construct a composite field. For GF((2 5 ) 2 ), we can choose f5(x)=x 5 +x 2 +1 to construct the subfield GF(2) on GF(2) 5 ), and choose f2(x)=x 2 +x+1 in GF(2 5 ) construct GF((2 5 ) 2 ). Figure 7 Shows the relationship between the various fields.

[0051] Although this is not the case here, we note that if n is 2 and m is even, we cannot use the irreducible polynomial f2(x)=x 2 +x+1 to GF(2 m ) construct GF((2 m ) 2 ). Instead, we can calculate the tracking value Tr(θ) for various values ​​of ii ), find the form f2(x)=x 2 +x+θ i Irreducible polynomials of θ, where GF(2 m ) in the original element. If Tr(θ i )=1,f2(x)=x 2 +x+θ i For GF(2 m ), but otherwise reducible. We can choose a i The irreducible polynomial with the lowest Hamming weight value is in GF(2 m ) construct GF((2 m ) 2 ).

[0052] Since we are considering the 5 ) we notice that there exists a n ) to perform efficient solution of general multiplication of Hamming weight polynomials (such as trinomials or quinomials) on the system Verilog. The System Verilog function is:

[0053] Therefore, according to the above method, the GF(2 5 ) where N=5.

[0054] Now we introduce multiplication in composite field multiplication. Let β be the irreducible polynomial f2(x)=x 2 +x+1, that is, β 2 =β+1. Let a, b, c∈GF((2 5 ) 2 ), c = a·b, where a = a1β+a0, b = b1β+b0, c = c1β+c0 and a0, a1, b0, b1, c0, and c1∈GF(2 5 ). We can derive the equations for calculating the coefficients c0 and c1: c=c1β+c0=(a1β+a0)(b1β+b0)=a1b1β 2 +(a0b1+a1b0)β+a0b0 c=(a1b1+a0b1+a1b0)β+(a1b1+a0b0) (1) Through Karatsuba's method, c=((a0+a1)(b0+b1)+a0b0)β+(a1b1+a0b0), c1=(a0+a1)(b0+b1)+a0b0, c0=a1b1+a0b0 (2)

[0055] For GF(2 n ), we can use the following equation to calculate the square of the field elements. Let a∈GF(2 n ),and where a i ∈GF(2), and α is GF(2 n ) is a polynomial basis generator for . Next, we use the field generator polynomial to perform the reduction. 5 +x 2 +1Construct GF(2 5 ). Let a=a4α 4 +a3α 3 +a2α 2 +a1α+a0, and c=a 2 =a4α 8 +a3α 6 +a2α 4 +a1α 2 +a0. By α 8 =α 5 α 3 =(α 2 +1)·α 3 =α 3 +α 2 +1 and α 6 =α 3 +α, we have c=a2α 4 +(a4+a3)α 3 +(a4+a1)α 2 +a3α+(a4+a0) For the square of the composite field elements, where c, a∈GF((2 5 ) 2 ), and c0, c1, a0 and a1∈GF(2 5 ).

[0056] Turning now to square root operations in composite fields, we first review the calculations in GF(2 5) is an algorithm for finding the square roots of non-zero elements in . Let a, c∈GF(2 5 )and c=a 2 We use the vector form [a0, a1, a2, a3, a4] and [c0, c1, c2, c3, c4] to represent a and c, where a0~a4 and c0~c4∈GF(2). GF(2 5 ) can be viewed as a 1×5 vector multiplied by a 5×5 matrix, that is, [c0, c1, c2, c3, c4] = [a0, a1, a2, a3, a4] · A, where Any element has only one square root, that is, the matrix A can be inverted as Through its [a0, a1, a2, a3, a4] = [c0, c1, c2, c3, c4]·A -1 Now we derive the calculation of GF((2 5 ) 2 ) is the square root of the non-zero elements in . From Equation 3, we know that: Among them, c0, c1, a0 and a1∈GF(2 5 ).

[0057] We now proceed to show how to perform multiplicative inverse operations on composite fields. n ), any non-zero field element can be expressed as a power of the original element, i.e., a = α i .pass A 31-depth LUT can be used to implement GF(2 5 ) in the multiplication inverse operation. For the composite field GF((2 5 ) 2 ), let a, b∈GF((2 5 ) 2 ), a=a1β+a0,b=b1β+b0,where a0,a1,b0 and b1∈GF(2 5 ). a0 and a1 cannot be zero at the same time. By a·b=1, we can get two linear equations for b0 and b1: a1b0+(a0+a1)b1=0 (5) a0b0+a1b1=1 (6) From equations 5 and 6, we can see that: Therefore, the composite field GF((2 5 ) 2 ) can be reversed by a set of GF(2 5 ) multiplier, squarer and a GF(2 5 ) inverter (which can be implemented via a 31-deep LUT).

[0058] With the above groundwork done, we can efficiently determine the error location polynomial. We note that the (1022, 990) extended BCH code has a correction power of 3, and rather than relying on the Chien search algorithm to factorize the polynomial, which would impose a delay proportional to the code size, we limit the degree of the error location polynomial to 3. Fast factorization can be achieved by solving cubic, quadratic, or linear polynomial equations in the composite field.

[0059] We start by solving a quadratic equation. The equation x 2 =a, where a∈GF(2 n ), always in a∈GF(2 n ) has a double root. For a received message with two errors, the error locator polynomial will take the form x 2 +ax+b=0 form, where a, b∈GF(2 n ), α≠0. This form can be converted to the following equation by replacing x=a·t set up Equation 9 has two roots, r and r+1, if and only if in is from GF(2 n ) to GF(2) is mapped as follows:

[0060] Before showing how to solve Eq. 9 in the composite field, we examine GF(2 10 ), which is obtained by p(x)=x 10 +x 3 +1 constructed. GF(2 10 ) elements s can be obtained through the polynomial basis {1, α, α 2 , ..., α 9 ) is expressed as where α is a root of p(x) and s i ∈GF(2). It can be shown that for all i≠7, is zero, which means

[0061] If GF(2 10) in which there exists an equation x 2 +x+s=0, then By using the equation Tr(x 2 +x+s)=Tr(x 2 )+Tr(x)+Tr(s)=s7=0 on both sides (apply tracking mapping Abbreviated as Tr), because Tr(x 2 )=Tr(x). For GF(2 10 ) elements, we only need to check the coefficients to determine whether the quadratic equation is in GF(2 10 ) has roots in . (Incidentally, we note that this technique works for all binary fields GF(2 n )).

[0062] When the coefficient is zero, let x1 be a root. Then the other root is x2=x1+1. By Hilbert construction method: where θ∈GF(2 10 ) and Tr(θ) = 1. By choosing θ = α 7 : These The terms are constants that can be precomputed, the addition is a bitwise XOR operation, and its squaring can be performed by the XOR array in a manner similar to the polynomial multiplication Verilog code above. These calculations can be expressed as multiplications of the coefficient vector and a 10x10 matrix, whose entries are GF(2) elements.

[0063] The underlying field is the composite field GF((2 5 ) 2 ), s∈GF((2 5 ) 2 ), and it can be expressed as s = s1β + s0, where s0, s1∈GF(2 5 ). A root is represented as c=c1β+c0 by replacing β with β+1. 2 , we have the equation Equations 14 and 15 are subfields of GF(2 5 ) to make them easier to solve. Equation 14 can either be expressed in GF(2 5 ) has two different roots, either in GF(2 5 ) has no roots, if Equation 14 is in GF(25 ), then there will be no roots in the composite field of Equation 9. Assuming that roots exist, one root is r and the other is r+1.

[0064] Replace c1 with r and r+1 in Equation 15. and If one of is zero, the other must be 1. Keep the root to satisfy Given the selected roots, we can also find the two roots c0 of equation 12 given the selected root c1.

[0065] A depth-31 LUT can be used to determine r, and therefore r+1, for Equations 14 and 15, but a simpler approach can be used to calculate the value of the LUT in GF(2 5 ) to solve the quadratic equation x 2 +x+s=0 (this method is based on Hilbert's theorem 90)

[0066] In GF(2 5 ) in the equation x 2 +x+s=0 has a root, so Tr(s)=0. Proof: By applying the tracking map on both sides of the equation, Tr(x 2 +x+s)=Tr(x 2 )+Tr(x)+Tr(s)=0 By Tr(x 2 )=Tr(x), Tr(s)=0. Because Tr(s)=0 and in GF(2 5 )middle but is one root of the equation, and the other root is Using the above foundation, we can use the arrangement of logic gates to find the roots. We do not need any LUT to solve GF(2 5 ) and GF((2 5 ) 2 ) 2 +x+s=0. Once these roots are found, the variable substitution of Equation 9 can be reversed to determine the roots of the quadratic error position polynomial. These calculations can be expressed as multiplications of the coefficient vector and a 10×10 matrix whose entries are GF(2) elements.

[0067] When choosing an irreducible polynomial f5(x) = x 5 +x 2 +1 to construct GF(2 5 ), it is easy to trace back the existence of a root. Let α be a root of f5(x). Then the field elements, α, can be obtained by the polynomial basis {1, α, α 2 , α 3 , α 4} is represented as Because Tr(α)=Tr(α 2 )=Tr(α 4 )=0 and Tr(1)=Tr(α 3 )=1, Tr(a)=a0+a3.

[0068] Now go to the composite field GF((2 5 ) 2 ), we only need to consider two types of cubic equations: 3 +d=0, where d≠0, and x 3 +x+d=0. This simplification comes from the general cubic equation, x 3 +ax 2 +bx+c=0,a≠0,a,b,c∈GF((2 5 ) 2 ) can be converted to these two types. The conversion is done by variable substitution, replacing x with t+a, thus producing t 3 +(a 2 +b)t+(ba+c)=0.

[0069] If a 2 + b = 0, we have converted the general cubic equation to the first type. Otherwise, when a 2 +b≠0, replace t with sy, and produce s 3 y 3 +(a 2 +b)sy+(ba+c)=0. divide it by s 3 To obtain set up Make Then we have the second cubic equation of type y 3 +y+d=0.

[0070] For the first type, we can either find 3 different roots or GF((2 5 ) 2 ), if we find a root r, the other two roots will be rβ and rβ 2 , where β 2 +β+1=0,(β 3 =1). Assuming r = r1β + r0, we know rβ=r1β 2 +r0β=(r0+r1)β+r1 (16) rβ 2 = r1β 2 +r0β=(r0+r1)β 2 +r1β=r0β+(r0+r1) (17) We can use a LUT to save only one root, allowing the use of a LUT with a depth of only 341. The other two roots can be calculated by an arrangement of logic gates via equations 16 and 17. The logic gate arrangement can then reverse the variable substitution to provide the roots of the original cubic equation.

[0071] In GF((2 5 ) 2 ), for the second type of cubic equation x 3 +x+d=0: . The value of d is 512, where the equation is in GF((2 5 ) 2 ) has only one root, The value of d is 170, where the equation has 3 different roots, The value of d is 341, where the equation is in GF((2 5 ) 2 ) has no root, and For d=0, 1 is the second root and 0 is the third root. In our use case, the error locator polynomial is cubic only if there are errors in three different locations. Therefore, we only care about the value of d being 170 where the cubic equation has 3 different roots. The LUT only needs to store two roots for each of these values ​​since the third root is the sum of the first two. Once the roots are obtained, the variable substitution is reversed through an arrangement of logic gates to obtain the roots of the original equation.

[0072] After covering techniques for efficiently finding roots of quadratic and cubic equations, we move on to the implementation of the decoder, starting with the syndrome calculator 602.

[0073] Although the composite field GF((2 5 ) 2 ) and GF(2 10 ), but the H matrix defined in the ITU standard is not suitable for syndrome calculations. Instead, we should use H comp For syndrome calculations. comp We precompute the composite field GF((2 5 ) 2 ) to find the primitive polynomial p(x)=x defined in the ITU standard 10 +x 3 +1, which can be done by exhaustive enumeration using the composite field multiplication defined above. There are 10 conjugate roots that can be found. For simplicity, we choose the root α with the lowest Hamming weight, which is the root of p(x). In Verilog codewords, this is 10'h0C1. Using the composite field representation of α, we can calculate α i , α3i and α 5i , and H comp Each column of comp The last two rows of are the same as the ITU standard H matrix. We can then calculate the syndrome of the received word via the following equation, in represents the binary bit stream of the received word. We can choose a partially parallel architecture or a fully parallel architecture to implement it. As an example of a partially parallel architecture, we can scan 256 bits per cycle (254 bits in the last cycle) for 4 cycles, accumulating the partial sums each cycle into the syndrome register. For a fully parallel architecture, the syndrome calculator requires 1022 bits of input. If it is difficult to complete such a large XOR sum in one cycle, we can insert 2 or 3 pipeline stage registers to shorten the critical path.

[0074] Since the error correction capability of the (1022, 990) extended BCH code is only 3, the position finder 604 does not need to use the Berlekamp-Massey algorithm to find the error location polynomial, nor does it need to use the Chien search algorithm to decompose the error location polynomial. Instead, the error location polynomial can be calculated based on the syndrome. Once we find the error location polynomial, we can solve the above linear, quadratic or cubic equation to decompose the error location polynomial to determine the error location.

[0075] The syndrome calculator 602 provides the 32-bit syndrome vector to the location finder 604. The first 30 bits are α i , α 3i and α 5i The three GFs of the receiving polynomial ((2 5 ) 2 ) Evaluation result. The last 2 bits are used for additional parity to reduce the error correction rate. With an error correction capability of 3, there are four possibilities to consider for the number of errors in the received message: 0, 1, 2, or 3. The error corrector can determine which of the four possibilities applies based on the syndrome.

[0076] Improving the approach outlined by Truhachev et al., “Efficient FEC for 400 Gbps Optical Communications,” IEEE Transmission Circuits and Systems, v68n1, January 2021, we first define the error location polynomial as follows, which is the polynomial defined in most textbooks. Let eloc(X) = ∏(X + X i ),in i=1, 2 or 3. Case 1: If the first 30 syndrome bits are zero and either of the last 2 syndrome bits is non-zero, the received message has more errors than can be corrected. If all 32 syndrome bits are zero, the received message contains no errors and the output data is the same as the input. Case 2: If the received message contains a single error eloc(X)=X+X1, where the first syndrome bit S1=X1 and the twentieth syndrome bit and Syndrome 30 Bit Because X1≠0, S1≠0. Add the cube of S1 to S3, Add S1 to the fifth power to S5, If S1 is non-zero, and D3 and D5 are zero, then there is a single error at k1. k1 can be found by taking log(X1), or finding H whose first ten bits match X1. comp Column to find. Case 3: If the received message contains two errors, because X1≠X3→S1≠0.X1X2≠0→D3≠0. Therefore, if S1 is non-zero, and D3 is non-zero, and D5S1 = D3S3, then there are two errors. The coefficients of the quadratic error polynomial are 1, S1, and D3 / S1, and can be decomposed as described above. Case 4: If the received message contains three errors X1, X2, X3≠0, X1≠X2, X2≠X1≠X3. The error location polynomial is eloc(X)=(X+X1)(X+X2)(X+X3) =X 3 +(X1+X2+X3)X 2 +(X1X2+X2X3+X1X3)X+X1X2X3 Replace X with X1, X2 or X3, By summing the above three equations: The weighted sum from equations 19-21 We get Next, we will show the matrix, It is reversible. That is, we need to show that D3≠0. First, show that D3=(X1+X2)(X2+X3)(X1+X3) By equations (22) and (24), D3 = (X1 + X2) (X2 + X3) (X1 + X3) ≠ 0, Therefore, if S1 is non-zero, D3 is non-zero, and S3D3+S1D5 is non-zero, there are two errors. The coefficients of the quadratic error polynomial are 1, S1, and and The decomposition can be performed as described above.

[0077] In summary, the conditions for the above 4 situations are: cond1:S1=0&&S3=0&&S5=0; cond2:S1≠0&&D3=0&&D5=0; cond3:S1≠0&&D3≠0&&D5S1=D3S3; cond4:D3≠0&&S3D3+S1D5≠0; So far, we have shown how to compute the error location polynomial from three 10-bit syndromes. By the method proposed in Section 2, if we find that the number of roots is less than the degree of the error location polynomial, we can prove that the received word cannot be decoded.

[0078] Considering the above, Figure 8 is a flow chart of an illustrative decoding method that enables a BCH decoder to be implemented with an application-specific arrangement of logic gates and two relatively small lookup tables. Alternative implementations may include programmable logic devices, field programmable gate arrays, embedded controllers or processors executing suitable firmware, and general purpose processors executing suitable software from memory.

[0079] Block 802 represents the determination of syndrome values ​​S1, S3, S5, P by syndrome calculator 602. This determination corresponds to the H of the received message vector comp Matrix multiplication, the H compThe matrix can be implemented by a series of logic gates (AND gates for multiplication and XOR gates for summing the products). Although shown as depending on the test described later, blocks 812, 820 and 838 can be speculatively performed in advance by syndrome calculator 602, and addition, multiplication, power, root and inverse multiplication operations can also be performed via a series of logic gates to determine the values ​​of the relevant variables.

[0080] Block 804 represents a test of the location finder 604 to determine if the syndrome values ​​S1, S3, S5 are all zero, which corresponds to the absence of correctable errors. Although shown as either, blocks 806, 814, 822, and 832 may be performed by the location finder 604 in parallel with block 804 to determine which error location procedure should be performed. If the syndrome values ​​S1, S3, S5 are all zero, the location finder 604 checks if the parity value P in block 806 is zero. If not, the received message vector is marked as having uncorrectable errors in block 808. Regardless of whether there are no errors or the errors are uncorrectable, the received message vector is generated as output data in block 810.

[0081] Box 812 represents the value and Block 814 represents a test to see if these values ​​are all zero, which corresponds to the case where the received message contains a single error. If so, block 816 represents the location finder determining the single error location from the S1 syndrome value. As with the location corresponding to any polynomial root, the indexing circuit can determine the single error location by replacing H com The first 10 bits of each column of H are compared to the root to determine the bit index of the error in the received message vector. If the root is equal to the first 10 bits, then the column index (equal to the bit index of the received message vector) is the error location. Parity verification can be performed by accumulating H at the error location. com The last two bits of the column are used to calculate the parity delta syndrome. If the cumulative parity delta syndrome of all error positions is equal to the syndrome value P, the error is correctable.

[0082] Block 818 represents correcting the located error(s) by inverting the received message bits at each error location. Although not explicitly shown here, such correction may depend on determining that the located error is correctable.

[0083] Block 820 represents the calculation of the product values ​​D5S1 and D3S3. Block 822 represents a test of whether these values ​​are equal, corresponding to the case where the received message contains two errors. If so, block 824 represents the location finder determining the coefficients of the quadratic error locator polynomial and using variable substitution to obtain the trailing coefficients of Equation 9 Using the principles described previously, a logic gate array can be used to obtain the product of the combined value vector with a quadratic solution matrix to obtain two roots corresponding to the two error locations. The solution matrix can contain the inverse of the variable substitution used to obtain Equation 9. Block 826 represents the use of the root to find the matching H com An index circuit of a column index that is equal to a bit index of the received message vector, where the error may be found and corrected in block 818 if the error is determined to be correctable.

[0084] If all results are tested yes or no at blocks 804, 814, and 822, then the received message vector contains at least three errors. Block 832 represents a test by the location finder 604 to determine if the variable substituted cubic polynomial is a cubic of the first type, in which case the trailing coefficient value is In block 834, the location finder uses the depth-341 lookup table to get one of the three roots and uses it to compute the other two roots. Supporting logic gates can be used to reverse the variable substitution, i.e., add S1 to each root. Block 836 shows the use of the root to find the matching H com An index circuit of a column index that is equal to a bit index of the received message vector, where the error may be found and corrected in block 818 if the error is determined to be correctable.

[0085] If block 832 tests no, then the variable replaces the cubic to the second type. Block 838 represents the trailing coefficient value In block 840, the location finder uses a depth-170 lookup table to retrieve two of the three roots and combines them to compute the third root. Supporting logic gates can be used to invert the variable substitution, i.e., multiply each of them by And add S1. Block 842 represents using the root to find a matching H com A column index equal to a bit index of the received message vector where the error may be found and corrected in block 818 if the error is determined to be correctable.

[0086] Although the operations are described in order, it should be understood that these operations can be reordered, parallelized and / or combined. For example, block 824 can be implemented as a logic gate array to obtain two polynomial roots as the product of the syndrome value vector and the quadratic solution matrix. The quadratic solution matrix can be combined with operations corresponding to the determination of the quadratic equation trailing coefficient value s, the determination of the roots of the quadratic equation, and the inversion of variable substitution. Any sequence of operations that can be expressed as a linear transformation can be combined into a single combined linear transformation.

[0087] Each of the BCH decoder components (syndrome calculator, error location finder, and error correction logic) can be pipelined. The syndrome calculator operation is used to calculate the product of a 1x1022 vector and a 1022x32 matrix. If all 1022 input bits are available, the calculation can be performed via an array of AND-XOR gates. The XOR gates can be organized in a binary tree. A contemplated implementation uses 3 pipeline stage registers to shorten the critical path. A contemplated implementation of the location finder, including the error location polynomial calculation and the decomposition of the error location polynomial, is pipelined by 4 stage registers. If the number of roots found in the underlying field is less than the degree of the error location polynomial, the locator designates the error as uncorrectable. Otherwise, the error correction logic will perform further checks on the parity bits (the last two extended syndrome bits) and accumulate the check matrix H com The last two bits of the selected error position column in the decoder are checked to obtain the delta syndrome. The error correction logic requires 4 cycles to complete this further check. If the decoder is parallel, that is, all 1022 output bits are sent out at the same time, then the error correction logic only needs one cycle to correct the error bit. Therefore, the decoder latency is 12 cycles.

[0088] This BCH decoder design was synthesized using a commercial technology library. The target frequency was chosen to be 937.5MHz with a 20% margin. The synthesis tool used was the Outline Design Compiler. The total cell area of ​​the BCH decoder is 9311um^2. The area of ​​the standard voltage threshold (SVT) device is 91%, the area of ​​the low voltage threshold (LVT) device is 7.54%, and the area of ​​the ultra-low voltage threshold (ULVT) device is 1.45%.

[0089] Numerous alternatives, equivalents, and modifications will become apparent to those skilled in the art once the above disclosure is fully appreciated. It is intended that the claims be interpreted as covering all such alternatives, equivalents, and modifications that are included within the scope of the appended claims.

Claims

1. A circuit for correcting bit errors in a received version of a RS or BCH coded bit stream, the circuit comprising: A syndrome calculator having a first logic gate array for obtaining a syndrome value as a product of a received message vector and a parity check matrix, wherein the syndrome value comprises at least a first ten-bit syndrome value S1, a twentieth-bit syndrome value S2, and a thirtieth-bit syndrome value S3; a location finder for deriving an error number from the syndrome value, the location finder having: a second logic gate array for obtaining two polynomial roots as a product of a syndrome value vector and a quadratic solution matrix corresponding to a determination result of a trailing coefficient value s of a quadratic equation, a determination result of a root of the quadratic equation, and an inversion of variable substitution when the error number is 2; as well as indexing circuitry for determining a bit index for each of the polynomial roots; as well as An error corrector is configured to receive, for each received message vector, a set of zero or more bit indices representing error locations in the received message vector.

2. The circuit according to claim 1, characterized in that The inverse of variable substitution is denoted as x=S1·t, and the trailing coefficient values ​​are denoted as 3. The circuit according to claim 1, characterized in that The first logic gate array and the second logic gate array respectively include a set of logic AND gates and a set of logic XOR gates, the set of logic AND gates is used to implement bit-by-bit multiplication, and the set of logic XOR gates is used to implement bit-by-bit addition.

4. The circuit according to claim 1, characterized in that The syndrome value comprises an error parity value P, wherein the error corrector is configured to accumulate parity increments of the parity check matrix based on the bit index, and configured to invert a received message vector bit corresponding to the bit index if the parity increment indicates that the bit error is correctable.

5. The circuit according to claim 1, characterized in that The location finder further comprises: A first lookup table having supporting logic gates is used for the error number to be three and When three polynomial roots are obtained, the three polynomial roots correspond to the cubic equation x 3 + the root of d, where d is represented by And the depth of the first lookup table does not exceed 341.

6. The system according to claim 5, characterized in that The first lookup table contains a single polynomial root or error location, and the support logic gates derive the remaining polynomial roots or error locations from the single polynomial root or error location.

7. The circuit according to claim 5, characterized in that The location finder further comprises: A second lookup table with supporting logic gates is used to When three polynomial roots are obtained, the three polynomial roots correspond to the cubic equation x 3 +x+d, where d is represented by And the depth of the first lookup table does not exceed 170.

8. The circuit according to claim 7, characterized in that The second lookup table contains two polynomial roots or error positions, and the support logic gates of the second lookup table derive a remaining polynomial root or error position from the two polynomial roots or error positions contained in the second lookup table.

9. The circuit according to claim 1, characterized in that The position finder is configured to provide a syndrome value S1 as a polynomial root when the error number is 1.

10. The circuit according to claim 1, characterized in that The received message vector contains exactly 1022 bits, and the parity check matrix H comp It has 32x1022 binary elements, representing the composite field GF((2 5 ) 2 ) is the power of the root of the original polynomial in .

11. An error correction method, comprising: Obtain a syndrome value corresponding to the product of the received message vector and the parity check matrix, wherein the syndrome value includes at least a first ten-bit syndrome value S1, a twentieth-bit syndrome value S2, and a thirtieth-bit syndrome value S3; Converting the syndrome value to a set of zero or more polynomial roots corresponding to error locations in the received message vector, the converting comprising: deriving an error number from the syndrome value; and When the error number is 2, using a logic gate array to obtain two polynomial roots corresponding to the product of the syndrome value vector and a quadratic solution matrix corresponding to the determination result of the trailing coefficient value s of the quadratic equation, the determination result of the root of the quadratic equation, and the inversion of the variable substitution; and Determine the bit index of each polynomial root in the set.

12. The error correction method according to claim 11, characterized in that: The inverse of variable substitution is denoted as x=S1·t, and the trailing coefficient values ​​are denoted as 13. The error correction method according to claim 11, characterized in that: The logic gate array includes a set of logic AND gates and a set of logic XOR gates, the set of logic AND gates is used to implement bit-by-bit multiplication, and the set of logic XOR gates is used to implement bit-by-bit addition.

14. The error correction method according to claim 11, wherein: The syndrome value comprises an error parity value P, wherein the method further comprises: accumulating parity increments of the parity check matrix based on the bit index; and If the parity delta indicates that the bit error is correctable, each received message vector bit corresponding to one of the bit indices is inverted.

15. The error correction method according to claim 11, characterized in that: The conversion further comprises: Using the first lookup table with supporting logic gates, the error number is three and When three polynomial roots are obtained, the three polynomial roots correspond to the cubic equation x 3 + the root of d, where d is represented by The depth of the first lookup table does not exceed 341.

16. The error correction method according to claim 15, characterized in that: The conversion further comprises: Using a second lookup table with supporting logic gates, the error number is three and When three polynomial roots are obtained, the three polynomial roots correspond to the cubic equation x 3 +x+d, where d is represented by And the depth of the first lookup table does not exceed 170.

17. The error correction method of claim 11, further comprising providing a syndrome value S1 as a polynomial root when the error number is 1.

18. The error correction method according to claim 11, wherein: The received message vector contains exactly 1022 bits, and the parity check matrix H comp It has 32×1022 binary elements, representing the composite field GF((2 5 ) 2 ) is the power of the root of the original polynomial in .

19. The error correction method according to claim 11, wherein: Deriving the error number includes: If the syndrome values ​​are all zero, then the error number is determined to be zero; If the syndrome value S1 is non-zero and the remaining syndrome values ​​are all zero, then the error number is determined to be 1; if Then the error number is determined to be 2; and Otherwise, the error number is determined to be 3.

20. The error correction method of claim 19, further comprising: If the error number is 2, verify whether the quadratic equation has a root, the verification comprising determining that the number of bits of the binary representation of the trailing coefficient s is s7=0.

21. An error correction method, comprising: Obtain a syndrome value corresponding to the product of the received message vector and the parity check matrix, wherein the syndrome value includes at least a first ten-bit syndrome value S1, a twentieth-bit syndrome value S2, and a thirtieth-bit syndrome value S3; Converting the syndrome value to a set of zero or more polynomial roots corresponding to error locations in the received message vector, the converting comprising: deriving an error number from the syndrome value; and Using the first lookup table with supporting logic gates, the error number is three and When three polynomial roots are obtained, the three polynomial roots correspond to the cubic equation x 3 + the root of d, where d is represented by The depth of the first lookup table does not exceed 341; and Determine the bit index of each polynomial root in the set.

22. The error correction method according to claim 21, characterized in that: The conversion further comprises: Using a second lookup table with supporting logic gates, the error number is three and When three polynomial roots are obtained, the three polynomial roots correspond to the cubic equation x 3 +x+d, where d is represented by And the depth of the first lookup table does not exceed 170.

23. The error correction method according to claim 21, characterized in that: The syndrome value comprises an error parity value P, wherein the method further comprises: accumulating parity increments of the parity check matrix based on the bit index; and Each received message vector bit corresponding to one of the bit indices is inverted if the parity delta indicates that the received message vector bits are correctable.

24. The error correction method according to claim 21, characterized in that: The received message vector contains exactly 1022 bits, and the parity check matrix H comp It has 32×1022 binary elements, representing the composite field GF((2 5 ) 2 ) is the power of the root of the original polynomial in .

Citation Information

Cited By

  • Method and system for automatically acquiring weighing value of balance

    CN119621003A

  • An Automatic Acquisition Method and System for Weighing Values of a Balance

    CN119621003B