Protograph quasicyclic polar codes and related low-density generator matrix families
QC polarity codes address the limitations of LDPC and polar codes by eliminating short cycles and enabling high-performance, low-latency decoding with reduced complexity, suitable for latency-constrained systems.
Patent Information
- Application Number
- JP2023572254
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- Priority Date
- 2021-02-22
- Filing Date
- 2021-11-30
- Publication Date
- 2024-11-11
- Estimated Expiration
- 2041-11-30
AI Technical Summary
Existing error correction codes, such as LDPC and polar codes, face challenges in achieving high performance while balancing computational complexity, latency, and code length, particularly in latency-constrained systems like IoT and M2M communications, with LDPC codes requiring long decoding times and polar codes having poor parallelizability and short cycle issues.
Introduce quasi-cyclic (QC) polarity codes based on protograph design, which eliminate short cycles through hill-climbing optimizations and protograph replacement, allowing for highly parallelizable decoding with low computational complexity and reduced latency.
QC polarity codes achieve high coding gain and throughput with low latency and complexity, comparable to longer codes, while maintaining computational efficiency, suitable for latency-constrained systems.
Smart Images

Figure 0007580636000032 
Figure 0007580636000033 
Figure 0007580636000034
Abstract
Description
[Technical field]
[0001] The present invention relates generally to the field of error correcting codes for data storage and data transmission, and more particularly to a method and system for protograph-based quasi-cyclic polar codes and their generalized low-density generator matrix family. [Background technology]
[0002] Capacity-based forward error correction (FEC) based on low-density parity-check (LDPC) codes has contributed greatly to improving the reliability of data storage, wireless data transmission systems, and optical data transmission systems. LDPC codes use a parity-check matrix (PCM) H of size M × N, where M and N represent the number of parity bits and the length of the codeword bits, respectively. The code rate is defined as the net information bits per codeword length, i.e., K / N, where the number of information bits is at most K = NM. For binary LDPC codes, the PCM consists of "0" or "1". For non-binary cases with Galois field size Q, the elements of the PCM are formed by integers from 0 to Q-1. PCM restricts the valid codewords such that the M parity equations for an N-bit codeword are all zero. That is, for binary codes, the modulus 2 arithmetic uses xH T =0([.] Tdenotes transpose), where x is a row vector representing an N-bit binary codeword. The PCM of an LDPC code can be represented as a factor graph or a bipartite graph, where the N coded bits are denoted by variable nodes (VN) and the M parity checks are denoted by check nodes (CN). If the (i,j)-th element of the PCM is nonzero, the i-th CN and the j-th VN are interconnected in the factor graph. Typically, LDPC codes can be decoded by parallelizable belief-propagation (BP) decoding, which propagates belief messages using certain rules in the VNs and CNs to correct potential errors given the channel output, for example in the form of log-likelihood ratios (LLRs). BP decoding uses iterative message passing, such as sum-product, delta-min and min-sum algorithms, following certain scheduling, such as hierarchical round-robin and flooding.
[0003] Protograph codes are introduced to design LDPC codes or accumulate-repeat-accumulate (ARA) codes, where 1′ in the PCM is replaced by a permutation matrix of size Q×Q, and 0′ is replaced by an all-zero matrix of size Q×Q. The permutation size Q is also called the lifting size. The actual factor graph is formed from the compact protograph by a lifting operation, where we replicate the same short graph Q times and permute edges across the replicated graphs. This is equivalent to the parallel use of Q-fold LDPC codes with combination by permuted message exchange. The total codeword length increases to QN. The most typical choice of permutation matrix is a cyclic matrix of weight 1 obtained by cyclically right-shifting a unit matrix of size Q×Q by s positions, where s is the cyclic shift value to be designed. In this case, the protograph LDPC code is called a quasi-cyclic (QC) LDPC code. QC LDPC codes have many practical advantages. For example, i) cyclic permutations are easy to implement in hardware. ii) the basic PCM is easier to design since its codeword length is shorter by a factor of Q. iii) protograph codes can become less dense as the lifting factor Q increases, even if the original PCM is dense. iv) short cycles in the factor graph can be easily increased by cyclic shift design. It has been shown that the maximum girth of a conventional regular QC LDPC code is 12, where girth is the length of the shortest cycle in the code graph. It is particularly important to maximize the code girth for BP decoding to work properly in looped graphs. For some irregular QC LDPC codes, girths larger than 12 have been shown to be feasible. Assuming that short cycles are eliminated by maximizing the girth, the performance of QC LDPC codes can be analyzed by the protograph extrinsic information transfer (P-EXIT) technique, which tracks mutual information across message paths in the protograph.Since the protograph is Q times more compact than the exact factor graph, the P-EXIT analysis is more efficient for QC LDPC codes. Summary of the Invention [Problem to be solved by the invention]
[0004] [Technical issues] In modem systems, QC LDPC codes, such as spatially-coupled (SC) LDPC codes, have been used to achieve error correction performance close to the Shannon limit. However, the pursuit of high LDPC performance has resulted in a significant increase in power consumption and circuit size. Therefore, a proper tradeoff between performance and computational complexity is more important for practical systems. In addition, modern high-performance LDPC codes usually require very long codeword lengths that cause long decoding latencies, whereas shorter FEC codes are preferable for latency-constrained systems such as Internet-of-Things (loT) applications and machine-to-machine (M2M) communications. Furthermore, LDPC codes are generally not suitable for efficient encoding because the code design is based on PCM optimization, whereas the codeword generation requires a corresponding generator matrix (GM) “G” of size N×K. The GM is usually calculated as GH T = 0, we need the Gaussian elimination method of PCM to generate an N-bit codeword x as x = uG from K-bit information data u.
[0005] In recent years, polar codes have attracted much attention as an alternative capacity-approaching code to LDPC codes when the block length is short. It has been found that polar codes can asymptotically achieve the channel capacity in any channel, even with a simple algorithm of successive cancellation (SC) decoding, due to the so-called polarization phenomenon. In addition to encoder design methods such as density evolution (DE), several decoder algorithms have been developed, including successive cancellation list (SCL) decoding and BP list decoding. For SCL decoding, polar codes are highly comparable to state-of-the-art LDPC codes. In particular, it has been found that the SCL decoding performance is highly dependent on the embedded cyclic-redundancy check (CRC) polynomial, and that the minimum Hamming distance of polar CRC concatenated codes can be increased by a properly designed CRC code. So far, various extension versions based on polar coding have also been proposed in the literature, such as non-binary polar codes, mixed kernel polar codes, irregular polar codes, concatenated polar codes, convolutional polar codes, and polar turbo product codes.
[0006]
number
[0007]
number
[0008] Because polar codes are based on GM rather than PCM, unlike LDPC codes, both encoding and decoding can be performed systematically without extra computation. The log-linear complexity of polar SCL decoding is a major disadvantage compared to the linear complexity of LDPC BP decoding for long codewords, but it proves to be advantageous when shorter block sizes are preferred to reduce the decoding latency. It has been demonstrated that polar decoding is more efficient than typical LDPC decoding for short block lengths. This has promoted polar codes as a strong candidate for latency-critical systems.
[0009] However, polar SCL decoding is not well suited for parallel implementation due to the nature of the sequential decoding algorithm, which results in longer latency. Although parallelizable BP decoding is available for polar decoding, there has been a long-standing problem that BP decoding does not work well for polar codes due to the inherently short perimeter of 4. Non-binary polar codes and polar positive codes can partially provide limited parallelism, but the computational complexity for encoding and decoding is significantly higher than that of regular binary polar codes. [Means for solving the problem]
[0010] Some embodiments of the present invention are based on the recognition that data communication and data storage require error control techniques to ensure continuous transmission. Polar coding has been demonstrated to achieve channel capacity by using successive elimination decoding as forward error correction. However, successive elimination decoding has the drawback of its poor performance and long latency to completion. Parallelizable belief propagation decoding also does not work well because the factor graph of polar codes has many short cycles. The method and system of the present invention provides a way to solve the above-mentioned problems by introducing a protograph lifting extension for the polar coding family such that highly parallelizable decoding is realized to achieve high coding gain and high throughput without increasing computational complexity and latency. The present invention allows iterative message passing to work well by eliminating short cycles through frozen bit allocation and hill climbing optimization of the protograph permutation. Some embodiments use quasi-cyclic (QC) permutation to achieve a highly parallelizable structure suitable for hardware implementation. Another embodiment further reduces the complexity, latency and number of short cycles using irregular pruning of proto-polarization units in the protograph, which allows the shallow polarization stage to outperform conventional deep polarization stages without protograph lifting. Some embodiments further generalize the generator matrix of any linear code with a multi-stage decomposition to construct a polar low-density generator matrix (LDGM) code family. Some embodiments enable low-latency windowed scheduling with staircase spatial coupling.
[0011] The present disclosure provides a new polar coding based on protograph design, in which fully parallel short polar codes are combined with computation-free permutation message exchange in intermediate polarization stages to simultaneously achieve high coding gain, low computational complexity, low decoding latency, and high throughput processing. The method and system of the present invention provides a novel family of protograph-based polar codes, called quasi-cyclic (QC) polar codes, with cyclic permutations in the proto-polarization units in some embodiments. With appropriate cyclic shift values, QC polar codes can eliminate short cycles in the code graph, which achieves a significant breakthrough towards solving the long-standing problem that BP decoding does not perform well for conventional polar codes. In QC polar codes, highly parallel short polar codes are combined to achieve performance comparable to longer polar codes while maintaining the same low computational complexity as the shorter polar codes.
[0012] The present disclosure is different from any conventional protograph code based on the lifting operation of PCM to design LDPC codes or Accumulate Repeat Accumulate (ARA) codes. Unlike LDPC codes, the method and system of the present invention generalize the concept of protograph to the lifting operation of GM to design low-density generator matrices (LDGMs). Since the code is based on GM, the encoding is not difficult, unlike conventional protograph LDPC codes. In addition, the present invention provides a multi-stage structure by lifting the GM of the polarization operation, thereby realizing a new concept of proto-polarization to achieve the polarization phenomenon to approach the channel capacity. For some embodiments of protograph LDGM codes, the advantages of structured encoding and decoding inherent to polar coding are achieved by realizing a protograph-based polar code. Specifically, the polarization unit used in conventional polar codes is replaced with a proto-polarization unit consisting of a proto-variable node and a proto-check node, and multiple bits are processed simultaneously by the lifting operation. Some embodiments use highly parallelizable QC-polarity codes replicated by lifting operations using cyclic permutations on the proto-polarization units as a special case of protographs, which allows hardware-friendly parallel encoding and decoding without increasing computational complexity but with increased freedom in code design.
[0013] QC polar codes can significantly reduce computational complexity along with the protograph lifting coefficient. Furthermore, the present invention can eliminate short cycles in protograph polar codes to increase their perimeter by hill climbing. Eliminating short cycles allows BP decoding to work well for QC polar codes. QC polar codes with shallow polarization stages can achieve competitive performance with deeper polarization codes. In some embodiments, further complexity reduction and improved performance is achieved by pruning proto-polarization units irregularly to break loops in the protograph.
[0014] Some embodiments are based on the recognition that the computational complexity and decoding latency of polar codes, unlike LDPC codes, are of the order of a log-linear function of the number of codeword bits. Thus, by shortening the length of the codeword, the required computational complexity as well as the decoding latency can be significantly reduced. In addition to cyclic message exchange, the present disclosure provides a method and system for using multiple short polarity encoders and decoders that are fully implemented in parallel without increasing complexity. This allows the short polarity codes to perform better than conventional short polarity codes while achieving the same low computational complexity and low decoding latency as shorter polarity codes.
[0015] Some embodiments are based on another realization that while polarity codes converge towards optimal coding performance close to the Shannon limit for deep polarization stages, their practical error correction performance can be improved for shorter code lengths by spatially combining multiple polarity codes. The method of the present invention allows shallow polarization to achieve performance comparable to that offered by deeper polarization without increasing complexity. Some embodiments are realized by spatial combination of non-uniform short polarity codes with different frozen bit positions. For example, spatial combination with a staircase lifting operation allows low latency windowed decoding for infinitely long codeword lengths for some embodiments.
[0016] In some embodiments, multiple cyclic redundancy check (CRC) codes are embedded across the spatially coupled encoder so that each bit plane can be determined independently as an incremental frozen bit. The method of the present invention accelerates the convergence of an iterative message passing algorithm by using incremental frozen bits. Some embodiments use probabilistic hard decisions based on the magnitude of the belief messages to create a list of hard decision candidates to improve the likelihood of a valid CRC codeword. With a high lifting factor, some embodiments may make the rate loss due to the compound CRC redundancy negligible.
[0017] Some embodiments are based on the realization that the perimeter of polar codes can be significantly increased by using an appropriate permutation matrix in the protograph, while conventional polar codes have an inherent perimeter of 4. Due to their short perimeter, conventional polar codes do not perform well with iterative decoding such as BP decoding, because message passing often gets trapped in short cycle loops in the code graph. The method of the present invention provides a method for eliminating short cycles by performing protograph lifting operations with permutations and cyclic shifts without the need for additional complexity using hill climbing. BP decoding of QC polar codes can be comparable to state-of-the-art SCL decoding. BP decoding provides highly parallel decoding, increasing the throughput of the decoding process.
[0018] Some embodiments are based on the recognition that protograph polarity codes can use techniques established for conventional QC LDPC design in a similar manner. Appropriate cyclic shift values for QC polarity codes can be designed with greedy and hill-climbing methods to increase the perimeter. Protograph polarity codes are compact compared to the entire factor graph, so code design can be more efficient. In addition, the present invention provides a method and system for designing frozen bit positions and irregular pruning and shift values by using P-EXIT evolution for finite iteration BP decoding. In addition, protograph polarity codes have the advantage of increasing the flexibility of codeword lengths beyond powers of two by adjusting the lifting coefficients. These protograph code specifications are adaptively controlled according to prior knowledge of channel statistics such as signal-to-noise ratio.
[0019] Some embodiments are based on the realization that the generator matrix (GM) of a polar code is decomposed into multi-stage polarization operations. The method of the present disclosure provides a method for decomposing any generator matrix of a linear block code into multiple simpler stages. For example, cyclic codes such as Hamming codes and BCH (Bose-Chaudhuri-Hocquenghem) codes can be systematically extended and decomposed into multiple polarization stages specified by corresponding polynomial coefficients. The decomposed multi-stage operations are further expressed as proto-polarization units to create a generalized LDGM protograph with high perimeter. The generalized LDGM protograph based on the linear coding GM is called a polar-type LDGM code because it has multiple polarization stages and frozen bits similar to the polar code. The LDGM protograph replicates the same multi-stage proto-polarization with lifting coefficients and permutes the polarization units between the replicated stages to provide a large-scale factor graph based on the small-scale protograph. This process of replication and permutation is called a lifting operation. Protograph LDGM codes have the advantage that encoding and decoding share the same graphical structure based on their GM without having to compute their PCM, while traditional LDPC protographs are only good for decoding based on their PCM, and their encoding requires increased complexity to generate their GM counterparts.
[0020] Some embodiments are based on the realization that adaptively scheduling message passing over the protograph reduces the required number of BP decoding iterations for protograph polar LDGM codes. The message passing schedule is designed based on P-EXIT evolution by greedy annealing optimization. The decoding convergence can be further accelerated by taking into account finite lifting coefficients as probabilistic mutual information updates. In one embodiment, the scheduling design is performed adaptively given the channel output data. Yet another embodiment uses a lookup table given the mutual information patterns to select an adaptive scheduling policy depending on the message passing rule and quantization precision.
[0021] Some embodiments are based on the realization that the adaptability of a polarity code to changes in parameters of a communication channel depends on the values of parameters such as a combination of parameters defining the number of data bits in a codeword, a parameter defining a data index set specifying the location of frozen bits in an encoded codeword, a parameter defining the number of parity bits in an encoded codeword, a parameter defining a polarization kernel, a parameter defining a polarization stage depth, a parameter defining a lifting size, a parameter defining a base generator matrix, a parameter defining a permutation in a proto-polarization unit, a parameter defining a spatial coupling, a parameter defining a selection of deactivated polarizations, a parameter defining a CRC embedding, a parameter defining a message passing rule, a parameter defining a decoding iteration strategy, a parameter defining an interleaver, a parameter defining a hard decision rule, etc.
[0022] These parameters are referred to herein as protograph specification parameters. A protograph code designed using the protograph specification parameters is referred to herein as a protograph polarity type LDGM code. An encoder that encodes source data using a protograph polarity type LDGM code is referred to herein as a protograph polarity encoder. A decoder that decodes noisy codewords encoded by a protograph polarity encoder is referred to herein as a protograph polarity decoder. A protograph polarity encoder (decoder) that uses a cyclic permutation matrix with weight 1 is referred to herein as a QC polarity encoder (decoder).
[0023] For example, some embodiments use polar code duplication, and the length of the codeword is relatively short, so that decoding can be performed with low latency and low complexity. In this case, each decoding corrects potential errors by propagating the belief message information back and forth. For this iterative polar encoding and decoding architecture, the error correction performance is improved by combining short polar codes, for example, with different code rates, different frozen bit positions, different codeword lengths, different interleaving, and different polarization kernels. In yet another embodiment, the polar encoding architecture is permuted between different polarization stages to achieve convolutional encoding and windowed decoding to reduce the decoding latency. Tail biting is used with or without zero padding for long convolutional protograph polar LDGM codes. To reduce the complexity and improve the performance, this protograph polar LDGM code architecture is further generalized by irregularly deactivating some proto-polarization operations. In some embodiments, the deactivation proto-polarization operation is further generalized by involving a variety of non-binary and higher order kernels and multiple weight permutations.
[0024] Accordingly, some embodiments disclose a method for reliably transferring data from a transmitter to a receiver over a communication channel. The method includes accepting source data and encoding the source data with a protograph polarity LDGM code to generate an encoded codeword. The protograph polarity LDGM code is specified by a set of parameters including a combination of a parameter defining the number of data bits in the codeword, a parameter defining the depth of the proto-polarization stage, a parameter defining a data index set specifying the location of frozen bits in the encoded codeword, a parameter defining the number of parity bits in the encoded codeword, a parameter defining a protograph lifting size, a parameter defining a base generator matrix with shift values, a parameter defining the randomness of the permutation of intermediate bits, a parameter defining the randomness of the polarization kernel in the polarity code, a parameter defining the randomness of the spatial coupling, and a parameter defining the randomness in the selection of deactivated proto-polarizations at various stages of polarity encoding. The protograph polarity LDGM encoder encodes the source data using the parameters of the code. The method further includes modulating the encoded codeword, transmitting the modulated message over a communication channel, receiving a noisy version of the modulated message, and decoding the demodulated codeword to reliably recover the original source data from the message passing.
[0025] Some embodiments disclose a system having a transmitter and receiver pair for reliably transmitting data over a communication channel, where the transmitter performs the above-mentioned method for reliably transmitting data over the communication channel, and the receiver performs the above-mentioned method for reliably retrieving the original source data by message passing.
[0026] According to some embodiments of the present invention, a computer-implemented encoding method for encoding digital data is provided. The encoding method can be performed by one or more computing processors to define a protograph structure of a polar-type low-density generator matrix (LDGM) code based on a code specification, the code specification including at least two stages of proto-polarization, at least one rule of frozen bit allocation, at least one rule of protograph substitution, and at least one rule of message passing. The method includes the steps of accessing a source bit stream as input digital data, initializing a data array with the source bit stream and a set of frozen bits according to the rules of frozen bit allocation, and propagating the data array according to the rules of message passing across all stages of proto-polarization. The propagating step further comprises (a) sequentially supplying the data sequence to proto-polarization stages from one stage to another, the proto-polarization stages including at least one proto-polarization unit, (b) permuting the data sequence in each of the proto-polarization units according to the protograph substitution rules, and (c) modifying the data sequence in each of the proto-polarization units according to the message passing rules. The computer-implemented encoding method further comprises arranging the modified data sequence into a codeword in a specified order, and providing the codeword as an encoded digital data of the input digital data.
[0027] The encoding method can increase robustness against unknown noise and errors that may occur through the communication channel. In addition, the robustness against unknown noise and errors can be further increased once the rules of frozen bit allocation and protograph substitution are determined by an iterative method that maximizes the code perimeter by eliminating short cycles in the protograph and optimizes the positions of frozen bits so that mutual information is maximized in the protograph extrinsic information transfer. The encoding method can solve the long-standing problem that conventional polar codes have inherent short cycles of length 4, which makes parallelizable decoding such as BP decoding unusable.
[0028] Yet another embodiment of the present invention provides a computer-implemented decoding method for decoding noisy codewords, the decoding method comprising the steps of: defining, on one or more computing processors, a protograph structure of a polar-type low-density generator matrix (LDGM) code based on a code specification, the code specification including at least two stages of proto-polarization, at least one rule of frozen bit allocation, at least one rule of protograph substitution, and at least one rule of message passing. The decoding method further comprises: accessing input data representing belief messages about noisy codewords; and initializing an array of left and right messages associated with each edge in the protograph. The right messages for a first stage of the proto-polarization are fixed based on the frozen bit allocation rules, and the left messages for a last stage of the proto-polarization are fixed based on the input data in a specified order. The decoding method further includes propagating the left and right messages according to a message passing rule across a plurality of stages of the proto-polarization including at least one variable node and at least one check node. The left and right messages are updated by an iterative method. The iterative method includes the steps of (a) selecting a parallelizable set of variable nodes and check nodes according to a message passing rule, (b) supplying the set of leftward messages and the set of rightward messages as incoming messages to the selected variable nodes and check nodes according to the protograph replacement rule, (c) computing a set of outgoing messages based on the incoming messages at the selected variable nodes and check nodes according to the message passing rule, (d) replacing the set of leftward messages and the set of rightward messages with the outgoing messages at the selected variable nodes and check nodes, and (e) repeating steps (a)-(d) a specified number of times according to the message passing rule. The decoding method further includes the steps of selecting the set of leftward messages as output messages in a specified order and providing the output messages as decoded digital data.
[0029] Thus, the decoding method can reduce the error rate by utilizing iterative message passing to correct unknown noise and errors. This decoding method of the present invention can solve the long-standing problem that conventional polarity codes are not amenable to parallel decoding due to their short period of four, which makes parallelizable decoding such as BP decoding unusable. The method can increase the decoding throughput without increasing the decoding complexity because it uses replicated short polarity codes. Thus, high encoding gains can be achieved with the same low complexity as conventional short polarity codes. In addition, message passing decoding allows for high speed decoding due to highly parallel decoding.
[0030] Some embodiments of the present invention provide an encoder circuit configured to generate encoded digital data from a source bitstream as input digital data. The encoder circuit can increase robustness against unknown noise and errors. The encoder includes an encoder interface configured to receive the source bitstream, at least one memory bank configured to store a computer-executable encoding method, and one or more computing processors configured to generate codewords from the input digital data as encoded digital data by executing steps of the computer-executable encoding method. The steps of the encoding method further include defining a protograph structure of a polar-type low-density generator matrix (LDGM) code based on a code specification, the code specification including at least two stages of proto-polarization, at least one rule of frozen bit allocation, at least one rule of protograph substitution, and at least one rule of message passing. The steps further include initializing a data array with the source bit stream and a set of frozen bits according to at least one rule of the frozen bit allocation, and propagating the data array according to at least one rule of the message passing throughout the stages of the proto-polarization. The propagating step further includes (a) sequentially supplying the data array from one stage to another to at least two stages of the proto-polarization, each stage of the proto-polarization including at least one proto-polarization unit, and further includes (b) permuting the data array in each of the proto-polarization units according to at least one rule of the protograph permutation, and (c) modifying the data array in each of the proto-polarization units according to at least one rule of the message passing. The steps of the encoding method further include generating the codeword as the encoded digital data based on a specified order by arranging the modified data array.For example, an encoder circuit may achieve a higher encoder throughput if the execution steps of a computer-executable encoding method are parallelized and serialized across a set of encoding methods by data exchange according to the rules of protograph substitution.
[0031] Further, some embodiments of the present invention provide a decoder circuit configured to generate decoded digital data from a noisy version of an encoded codeword, which can reduce unknown noise and errors. The decoder circuit includes a decoder interface configured to receive the noisy encoded codeword as input data, at least one memory bank configured to store a computer-executable decoding method, and one or more computing processors configured to generate the decoded digital data from the encoded codeword from the decoder interface by executing steps of the computer-executable decoding method. The steps of the decoding method include defining a protograph structure of a polarized low-density generator matrix (LDGM) code based on a code specification, the code specification including at least two stages of proto-polarization, at least one rule of frozen bit allocation, at least one rule of protograph substitution, and at least one rule of message passing. The steps of the decoding method further include initializing an array of leftward and rightward messages associated with each edge in the protograph structure, where the rightward messages for a first stage of the proto-polarization are fixed based on at least one rule of the frozen bit allocation and the leftward messages for a last stage of the proto-polarization are fixed in a specified order based on the encoded codeword. The steps of the decoding method further include propagating the leftward and rightward messages according to at least one rule of the message passing across the at least two stages of the proto-polarization including at least one variable node and at least one check node. The leftward and rightward messages are updated by an iterative method.The iterative method includes the steps of (a) selecting a parallelizable set of variable nodes and check nodes according to at least one rule of message passing, (b) supplying the set of leftward messages and the set of rightward messages as incoming messages to the selected variable nodes and check nodes according to at least one rule of the protograph substitution, (c) computing a set of outgoing messages based on the incoming messages at the selected variable nodes and check nodes according to at least one rule of the message passing, (d) replacing the set of leftward messages, leftward messages, and rightward messages with the set of outgoing messages at the selected variable nodes and check nodes, and (e) repeating steps (a)-(d) a specified number of times according to at least one rule of the message passing. The decoding method steps further include generating an output message as decoded digital data by selecting the set of leftward messages in a specified order.
[0032] Yet another embodiment of the present invention provides a system for transferring digital data over a communication channel. The system includes at least one computing processor, at least one memory bank, and a transmitter. The transmitter includes a source for accepting source data to be transferred and an encoder for encoding the source data, the encoder performing the computer-implemented encoding method to generate an encoded codeword. The transmitter further includes a modulator for converting the encoded codeword into a modulated signal, the modulated signal being based on amplitude modulation, phase modulation, quadrature amplitude modulation, phase shift keying, frequency modulation, orthogonal frequency division multiplexing, space division multiplexing, pulse position modulation, or variations thereof. The transmitter further includes a front-end interface for transmitting the modulated signal to the communication channel. The system further includes a communication channel, including a wired media channel, a wireless media channel, a fiber optic channel, a vacuum free space channel, a data storage medium, or variations thereof, and a receiver. The receiver includes a front-end interface for receiving an output signal from the communication channel, and a demodulator for demodulating the output signal of the channel. The demodulator generates likelihood information as an input belief message. The receiver further includes a decoder for decoding the input belief message, the decoder performing the computer-implemented decoding method to generate a decoded codeword. The receiver further includes a sink for providing the decoded codeword, and a controller for defining a code specification of a polar-type low-density generator matrix (LDGM) code to be used in the encoder and the decoder in response to knowledge of the channel. The code specification includes at least two stages of proto-polarization, at least one rule of the frozen bit allocation, at least one rule of protograph substitution, and at least one rule of message passing for encoding and decoding.
[0033] According to the system of the present invention, unknown noise and errors can be reduced. Furthermore, if each of the pipelined decoders further includes a set of parallel decoders with a set of permutation interconnects, the decoder throughput can be further increased. Furthermore, if each of the pipelined encoders further includes a set of parallel encoders with a set of permutation interconnects, the encoder throughput can be further increased. Thus, both the disclosed encoders and decoders can handle high throughput data, such as data center interconnects that require data throughput in excess of terabits per second. The disclosed system can achieve lower power data transmission using encoders and decoders for polar LDGM codes, since the implementation complexity remains as low as for short polar codes.
[0034] Yet another embodiment discloses a non-transitory computer readable storage medium having embodied thereon a program executable by a processor to perform the above-described method for securely transferring digital data. Effect of the Invention
[0035] Systems and methods for encoding protograph polarity codes and associated LDGM codes according to embodiments of the present disclosure can improve performance in error correction and decoding throughput, while also reducing computational complexity, decoding latency, and power consumption of real-time hardware processors. [Brief description of the drawings]
[0036] [Figure 1] FIG. 1 is a functional diagram illustrating a communication system for the transmission of digital data, according to some embodiments. [Diagram 2] 2 is a schematic diagram illustrating an example encoding employed by an encoder of the system according to some embodiments. [Diagram 3]FIG. 2 is a schematic diagram illustrating a polarization unit represented as a factor graph for processing message passing according to some embodiments. [Figure 4] 2 is a schematic diagram illustrating an example decoding employed by a decoder of the system according to some embodiments. [Diagram 5] FIG. 13 is a schematic diagram illustrating an example protograph lifting operation for a proto-polarization unit that propagates multiple bits simultaneously by performing multiple polarization unit duplication and edge substitution to combine parallel encoders, according to some embodiments. [Figure 6A] FIG. 13 is an exemplary diagram illustrating short cycles and cyclic shift values of a protograph polarity code for increasing the perimeter of a code graph, according to some embodiments. [Figure 6B] FIG. 13 is an exemplary diagram illustrating short cycles and cyclic shift values of a protograph polarity code for increasing the perimeter of a code graph, according to some embodiments. [Figure 7] FIG. 13 is a block diagram illustrating a perimeter maximization procedure for eliminating short cycles in a protograph by optimizing permutation shift values for a QC polarity code, in accordance with some embodiments. [Figure 8A] FIG. 1 is a schematic diagram illustrating an example of embedding one or more CRC codes into data bits for a protograph polarity code, according to some of the embodiments. [Figure 8B] FIG. 1 is a schematic diagram illustrating an example of embedding one or more CRC codes into data bits for a protograph polarity code, according to some of the embodiments. [Figure 8C] FIG. 1 is a schematic diagram illustrating an example of embedding one or more CRC codes into data bits for a protograph polarity code, according to some of the embodiments. [Figure 8D] FIG. 1 is a schematic diagram illustrating an example of embedding one or more CRC codes into data bits for a protograph polarity code, according to some of the embodiments. [Figure 9]1 is a block diagram illustrating a method for decoding a codeword using an embedded CRC code transmitted over a channel according to some embodiments. [Figure 10A] FIG. 1 is a schematic diagram illustrating bit-interleaved protograph polar coded modulation for non-uniform channels according to some embodiments. [Figure 10B] FIG. 1 is a block diagram illustrating a joint optimization procedure for an interleaver and a protograph polar code structure for a non-uniform channel, according to some embodiments. [Figure 11A] FIG. 1 illustrates an exemplary irregular QC polarity coding structure with sparsely selected inactive proto-polarization units, according to some embodiments. [Figure 11B] FIG. 13 illustrates an example irregular QC polarity coding structure with inactive protopolarization units for a two-stage QC polarity code, showing the effect of a deactivated protopolarizer to remove short cycles, according to some embodiments. [Figure 12] FIG. 1 is a block diagram illustrating a method for selecting an inactive protopolarizer of an irregular QC polarity coding structure, according to some embodiments. [Figure 13] FIG. 1 is an exemplary schematic diagram illustrating a method for spatially combining replicated polar encoders, according to some embodiments. [Figure 14] FIG. 2 is an exemplary schematic diagram illustrating a method for convolutionally combining replicated polar encoders, according to some embodiments. [Figure 15A] FIG. 2 is an exemplary schematic diagram illustrating a method for constructing a generalized polar-type LDGM code from a Hamming code, according to some embodiments. [Figure 15B] FIG. 2 is an exemplary schematic diagram illustrating a method for constructing a generalized polar-type LDGM code from a Hamming code, according to some embodiments. [Figure 16A]FIG. 2 illustrates an example of parallel encoding and decoding for a high throughput system according to some embodiments. [Figure 16B] FIG. 2 illustrates an example of parallel encoding and decoding for a high throughput system according to some embodiments. [Figure 17A] 1 is a block diagram illustrating a system suitable for implementing various components of a receiver for performing decoding in accordance with some embodiments and / or a transmitter for encoding a data message in accordance with some embodiments. [Figure 17B] FIG. 2 is a block diagram that illustrates generally an encoder circuit for generating encoded digital data from input digital data, in accordance with some embodiments of the present invention. [Figure 17C] FIG. 2 is a block diagram that illustrates generally a decoder circuit for decoding encoded digital data in accordance with some embodiments of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0037] [Description of the embodiment] Various embodiments of the present invention will now be described with reference to the figures. It should be noted that the figures are not drawn to scale, and that elements of similar structure or function are represented by similar reference numerals throughout the figures. It should also be noted that the figures are intended only to facilitate the description of certain embodiments of the present invention. They are not intended as an exhaustive description of the present invention or as limitations on the scope of the present invention. In addition, aspects described in connection with a specific embodiment of the present invention are not necessarily limited to that embodiment, and may be implemented in any other embodiment of the present invention.
[0038] In the field of digital communications, forward error correction (FEC) using an error correcting code (ECC) is a technique for encoding a message to add redundancy to mitigate the uncertainty introduced by a noisy communication channel, allowing a decoder to reduce transmission errors. In general, an ECC encoder is a system or method for converting a sequence of data symbols (representing a source message or source bit stream) into a more redundant sequence of code symbols (representing an encoded message or codeword) to be transmitted over a noisy channel. An ECC decoder is a system or method for recovering estimates of the original source data symbols from a noisy channel output that has been subjected to unknown random perturbations that must be corrected.
[0039]
number
[0040]
number
[0041]
number
[0042]
number
[0043] FIG. 1 illustrates a communication system for transmitting digital data from a transmitter 110 to a receiver 130 through a channel 120 according to some embodiments. For example, the communication channel 120 includes, but is not limited to, an air medium for wireless communication, a copper cable for wired communication, a solid-state drive for data storage transfer, a vacuum for free-space satellite communication, and a fiber cable for fiber optic communication. During communication for transferring digital data from the transmitter to the receiver, the digital data may be corrupted by noise such as AWGN that occurs across the channel and data processing path between the encoder 112 and the decoder 133. For example, the noise includes, but is not limited to, thermal noise, spontaneous emission amplifier noise, and impulsive shot noise. The transmitter 110 employs a forward error correction (FEC) code specified by the protograph parameters 140 to achieve reliable data transmission. The receiver employs the decoder 133 specified by the protograph parameters to recover the original data from the noisy message encoded by the encoder 112.
[0044] In the transmitter 110, data to be transmitted arrives from a source 111 configured to accept original data. The source may include, but is not limited to, a memory for storing data, an input port for receiving data, and / or a device for generating data. For example, in one embodiment, the source includes a voice communication device that converts an input voice signal into digital data. The input data from the source 111 is encoded by an FEC encoder 112. In some cases, the encoded data may be referred to as a codeword. The encoded data is modulated by a modulator 113 using various modulation formats, including, but not limited to, quadrature-amplitude modulation (QAM) with linear transformation such as orthogonal frequency-division multiplexing (OFDM) and QAM without such linear transformation. Modulation schemes include many other variations based on a combination of amplitude modulation, phase modulation, and frequency modulation, such as amplitude shift keying, phase shift keying, frequency shift keying, space division multiplexing, pulse position modulation, pulse amplitude modulation, etc. The modulated data is transmitted to a channel via a front-end circuit 114. The front-end circuitry 114 may include, but is not limited to, electro-optical circuits for optical communications and radio frequency circuits for wireless communications. The front-end may also include digital-to-analog converters and signal pre-processing, such as bandpass filters, pulse shaping, precoding, power loading, pilot insertion, and pre-distortion.
[0045] The channel 120 distorts the transmitted signal, for example, by introducing AWGN, co-channel interference, deep fading, impulsive noise, inter-symbol interference, Kerr-induced nonlinear interference, polarization crosstalk, and linear chromatic dispersion, as well as residual hardware imperfections such as quantization errors, clock jitter, overflow, laser linewidth, and carrier phase noise.
[0046] The receiver 130 first converts the channel output to an electrical received signal via a front-end circuit 131, which is typically complementary to the front-end 114 in the transmitter. For example, the front-end may include, but is not limited to, an analog-to-digital converter, linear equalization, nonlinear equalization, adaptive filtering, channel estimation, carrier phase recovery, synchronization, and polarization recovery. The received signal is demodulated in a demodulator 132 to generate initial estimates of the bits of the transmitted codeword that are used by a decoder 133 to recover the source data. For example, the demodulator provides log-likelihood ratio (LLR) values as soft input messages to the decoder 133. In various embodiments, the decoder 133 takes the soft input messages and generates soft output messages given a protograph code specification 140. Some embodiments are based on the recognition that the adaptability of a protograph code to changes in parameters of a communication channel depends on the values of the parameters, such as a parameter defining the number of data bits in a codeword, a parameter defining a data index set that specifies the location of frozen bits in the encoding data, and a parameter defining the number of parity bits in the encoded codeword. Thus, the protograph code specification 140 is adaptively controlled according to a priori knowledge 150 of the channel. For example, code parameters such as code rate and codeword length are adaptively selected based on channel knowledge of the signal-to-noise ratio (SNR), power delay profile, channel spectrum, and nonlinear statistics. In some embodiments, the decoder 133 is a hard-decision (HD) decoder for generating values indicative of the LLRs of bits based on the soft input and soft output messages of the decoder. The decoded data is provided to a data sink 134, which may include, but is not limited to, a memory for storing the data, an output port for transmitting the data, and / or a device for receiving the data. For example, in one embodiment, the sink includes a voice communication device that converts the decoded data into an audio signal.
[0047] The transmitter 110 and / or the receiver 130 may be implemented with a hardware processor operatively connected to a memory. Each of the transmitter 110 and the receiver 130 may include one or more processors. For example, the memory of the receiver 130 may store some information related to one or a combination of the protograph coding, the soft inputs and soft outputs of the decoder 133, the results of intermediate calculations, and the parameters of encoding and decoding. For example, the digital data is encoded with an encoding GM formed as a Kronecker power of a polar kernel matrix F. To that end, the memory of the receiver may store an encoding matrix used by the decoder's processor to decode the noisy codeword. All components in the transmitter 110 and the receiver 130, including the encoder 112, the modulator 113, the demodulator 132, and the decoder 133, may be implemented by hardware, one or more hardware processors, computer software (programs or program modules), or a combination of hardware and computer software. Polarity Encoding
[0048]
number
[0049] The same structure defined by the code specification 140 and used in the encoder 112 is used for the decoder 133. By applying the polarization operation 220 through multiple n stages 221, 222, 223, 224, the soft input message feeding the decoder can be gradually polarized into good and bad subchannels. The proportion of good subchannels with SC decoding asymptotically converges towards the Shannon limit of any channel for long codewords. This is called the polarization phenomenon. The polarization unit 220 has an upper branch and a lower branch, and the strides are often different from stage to stage. For example, the first polarization stage 221 uses a stride of 8, the second polarization stage 222 uses a stride of 4, the third polarization stage 223 uses a stride of 2, and the final stage 224 uses a stride of 1. This is because the GM of the polarized code can be decomposed into n sparse subgenerator matrices with off-diagonal elements at strides that are powers of 2 from the diagonal positions. For example, the GM for two-stage polarization is expressed as follows:
number
[0050] The first sub-GM with two off-diagonal elements corresponds to the first stage polarization with a stride of 2, and the last sub-GM corresponds to the second stage polarization with a stride of 1. Each stage has N / 2 off-diagonal elements corresponding to the XOR operation of the polarization units. This decomposed GM representation thus provides the definition of multi-stage connection rules for constructing the polarity coding protograph.
[0051] The polarization unit 220 uses XOR operation between the upper branch and the lower branch as shown in FIG. 3. The polarization unit 220 can be represented by a factor graph 310, which has an upper branch variable node (VN) 311, a lower branch VN 313, an upper branch check node (CN) 312, and a lower branch CN 314. The upper VN and the lower CN are of degree 2, i.e., the number of connected edges is 2. For those degree 2 factor nodes, they are often ignored because they do not involve computation operations. The remaining VN 313 and CN 312 are of degree 3, i.e., the number of connected edges is 3. One advantage of the polarity code is that its low-degree factor nodes are of degree 3 at most, and its computational complexity is the lowest for message passing. In contrast, typical irregular protograph LDPC codes use high-degree factor nodes, which results in high computational complexity in such high-degree nodes for message passing. Polarity Decoding
[0052] The decoder 133 is a computer-implemented method or system that receives (accesses) input belief messages from the demodulator 132 and transmits (provides) recovered bit sequences to the sink 134. The polarity decoder 133 processes the soft decision belief messages in each VN and CN of the polarization unit 220 using a specific rule of message passing 320. For example, the upper CN 312 takes the left LLR value A1 321 and the right LLR value B1 323 according to the check-node decoder (CND) rule E1=CND(A1,B1) to calculate the downward LLR value E1 330. The lower VN 313 takes the left LLR value A2 322 and the right LLR value B2 324 to calculate the upward LLR value E2 329 according to the variable-node decoder (VND) rule E2=VND(A2,B2). The leftward LLR values in the upper branch C1 325 and the lower branch C2 326 are then propagated through CN 312 and VN 313 to the next polarization stage as C1=CND(A1,E2) and C2=VND(A2,E1), respectively. Similarly, if necessary, the rightward LLR values in the upper branch D1 327 and the lower branch D2 328 are back-propagated through CN 312 and VN 313 to the previous polarization stage as D1=CND(B1,E2) and D2=VND(B2,E1), respectively.
[0053] The propagation rules for VND and CND include, but are not limited to, sum-product algorithms, delta-min algorithms, min-sum algorithms, offset min-sum algorithms, normalized min-sum algorithms, modified min-sum algorithms, decaying min-sums, extended min-sums, quantized lookup tables, or variations thereof. For example, the sum-product algorithm for soft-decision binary decoding has a message passing rule specified as follows: VND(A,B)=A+B, CND(A,B)=2tanh -1(tanh(A / 2)tanh(B / 2)), where tanh(.) is the hyperbolic tangent function, and tanh -1 (.) is its inverse function. In some embodiments, the belief messages are not LLR values but non-log likelihood ratios (LRs) ranging from 0 to ∞, or corresponding sigmoid probability regions ranging from 0 to 1. In another embodiment, the belief messages can be soft decision values that are direct noisy versions of the encoded codewords, or hard decision values to represent erroneous codewords. In some embodiments, the VND and CND rules are implemented by look-up tables (LUTs) to save computational complexity. In some embodiments, the belief messages are further quantized with low precision to save memory size of the LUTs. The quantization levels are adjusted per polarization unit according to the statistics. The decoder 133 may include a processor operatively connected to the memory and configured to perform the decoding step. Thus, the VND and CND rules are defined for the case of non-binary coding in some embodiments.
[0054] FIG. 4 shows an exemplary schematic diagram of a decoder 133 controlled by the protograph specification 140. The decoder structure in FIG. 4 is the same as the encoder structure in FIG. 2 in the sense that the same graphical representation of n-stage polarization is used for both encoding and decoding. This is a major difference between GM-based coding, such as LDGM codes, and PCM-based coding, such as LDPC codes, which requires GM computation in addition to PCM. For the message passing algorithm, all intermediate messages in each stage are initially initialized to zero, but the leftward message 410 in the rightmost stage (i.e., the last polarization stage) is set to the soft input message (e.g., the channel LLR value) provided by the demodulator 132, and the rightward message 400 in the leftmost stage (i.e., the first polarization stage) is set to either (0, +∞, -∞) depending on the freeze bit 201. Specifically, positions for frozen bits that are "0" have a prior belief message of +∞, positions for frozen bits that are "1" have a prior belief message of -∞, and otherwise for information data bits the prior belief message is 0. If the belief message is not in the LLR domain, the initialization is modified accordingly, e.g., a zero LLR value is equal to an LR of 1. Thus, the initialization of the belief message is defined according to the frozen bit allocation. A set of rules for propagating the belief message is then defined correspondingly for each proto-polarization unit.
[0055] The protograph code specification 140 provides scheduling parameters that determine which messages are propagated in what order across multiple VNs and CNs in the protograph. Some embodiments use round-robin scheduling that propagates rightward belief messages from the first stage to the second stage (420), from the second stage to the third stage (430), etc. Leftward belief messages are then propagated backwards from the last stage to the second last stage (430), from the penultimate stage to the penultimate third stage (431), etc. This round-trip message passing from left to right and right to left is repeated multiple times in an alternating fashion to correct potential errors. Within one stage, all messages can be updated simultaneously in parallel, or sequentially from top to bottom (440), or sequentially from bottom to top (450). In some embodiments, message passing is fully parallel in all VNs and CNs, regardless of stage as flooding scheduling to achieve highly parallel decoding. The final soft output decision is the leftward message in the first stage after iteration. The hard decision is based on the sign of the leftward message LLR value. The scheduling policy, message passing rules, and number of iterations are adaptively controlled by the protograph code specification 140 in some embodiments.
[0056] In some embodiments, a specific scheduling is used to sequentially obtain the leftmost message to realize SC decoding and SCL decoding, and the rightward LLR is always based on a hard decision message ranging from one of {0, +∞, -∞} depending on the availability of a hard decision based on the leftmost belief message. This SC / SCL decoding schedule further reduces the computational complexity, while the decoding latency increases overall. In some embodiments, a random scheduling or an adaptive scheduling for message passing is used depending on the instantaneous belief message, e.g., its magnitude or derivative.
[0057] It should be noted that the decoder 133 can be used as the encoder 112 with certain message passing scheduling and computation rules. Specifically, the right message is initialized in the first stage with encoding bits u, and the codeword x to be obtained can be generated by right message passing from the first stage to the last stage, where the message passing rule is based on XOR operations (or algebraic arithmetic operations in the non-binary case) at the protocheck nodes. Thus, some embodiments share the same modules for both encoding and decoding, where the message passing rules are controlled by the code specification 140. Protograph-based polarity code
[0058] The disclosed method and system provide a way to further improve the traditional polar coding and polar decoding originally used for LDPC or repeat-accumulate codes by introducing the concept of protographs. In 2003, Thorpe first introduced the concept of protograph codes as a class of LDPC codes constructed from protographs such that 1's in the PCM are replaced with a Q×Q permutation matrix and 0's in the PCM are replaced with a Q×Q zero matrix. The permutation size Q is also called the lifting size, lifting factor, or QC size. If the permutation matrix is cyclic, the protograph LDPC code is reduced to the well-known quasi-cyclic (QC) LDPC code. However, there is no prior art that uses protographs to design polar codes and related LDGM codes.
[0059]
number
[0060] Therefore, the encoder 112 shown in FIG. 2 and the decoder 133 shown in FIG. 4 have more specification parameters to realize a protograph-based polar code that includes a protograph lifting coefficient, i.e., an integer value Q. In a conventional polar code, n stages of polarization are N=2 n In the protograph polarity code of the present invention, the total number of codeword bits is N=2 n The number of information bits is also increased by a factor of Q (231). i and the encoded message x i is a Q-bit tuple for the protograph polarity code, rather than a single binary bit. Also, the XOR operation in each polarization unit is extended as proto-polarization units process Q bits in parallel. Protograph polarity codes have a number of superior features, including but not limited to, better performance than traditional polarity codes with deeper polarization stages, but with computational complexity as low as that of shallow polarizations.
[0061]
number
[0062] To achieve good performance, we shall design the permutation shift values of the QC polarity code. One obviously inferior choice would be for all shift values to be zero, resulting in Q-parallel short polarity codes that are independent of each other, without the benefit of combining. Conventional protograph LDPC codes are often designed to achieve high girths. The "girth" of a code is the length of the shortest cycle in the code graph. It is well known that the girth of any conventional QC LDPC code is upper bounded at 12. Tanner proposed a systematic method to optimize the shift values to achieve a girth of 12. Furthermore, it has been shown that irregular QC LDPC codes can achieve girths greater than 12.
[0063]
number
[0064] The factor graph of polar codes is loop-like in nature, and there are many short cycles in the code graph, which is why BP decoding cannot work well for conventional polar codes. Nevertheless, by optimizing the shift value, we can increase the perimeter for QC polar codes when Q>1.
[0065]
number
[0066] FIG. 6B shows another example of a short cycle for a three-stage QC polarity code having 12 proto-polarization units with shift values forming an elementary matrix of size 3×4 as follows:
number
[0067] This is s 13 →s 34 →s 24 →s 12 →s 31 →s 21 Cycle 6 loop 660 in the message path of . Similarly, this cycle 6 loop can be eliminated if the accumulated shift value is non-zero, as follows: -s 13 -s 34 +s 24 +s 12 +s 31 -s 21 ≠0(mod Q) Again, the shift value is negative for downward paths in loops of the chord graph. If both cycle 4 and cycle 6 loops are eliminated, the chord perimeter can be at least 8. In practice, eliminating cycle 4 is most effective in achieving good performance. Similarly, higher cycles with more polarization stages can also be eliminated. When the lifting factor is large, even the random shift base matrix can effectively eliminate short cycles overall. Note that the frozen bit positions can reduce the length of short cycles, whereas pruning of the proto-polarization can also eliminate some of the short cycles. Therefore, the optimization of the frozen bit allocation and the shift value are both important.
[0068] The disclosed method and system provides a method for designing high cyclic length shift values for QC polarity codes by using a greedy hill-climbing method to find cyclic shift values that eliminate short cycles in the protograph polarity code. Figure 7 shows a block diagram for designing high cyclic length QC polarity codes. First, a cyclic shift value of size n × 2 n-1The permutation shift values in the base matrix S of are initialized (700) by random integers evenly distributed between 0 and Q-1 for Q lifting coefficients, for example. We search the n-stage protograph polarity code to find (701) short cycle loops including cycle 4, cycle 6, cycle 8 and cycle 10 with stages of 2, 3, 4 and 5, respectively. Weighted scores are calculated for all shift values (702). The scores depend on the length of the cycle based on a weighting control (703). For example, if a shift value involves a cycle 4 loop, the score is increased by 100 points, and if a shift value involves a cycle 6 loop, the score is increased by 10 points. As mentioned above, if the accumulated shift value is non-zero, the score is not increased unless the score is increased by a weighting point. Note that two cycle 4 loops are equivalent to a cycle 8 loop. After calculating all the scores at each shift value, the maximum among the scores is compared to a target value (704). If the maximum score is below the target value, the optimization iteration is terminated and an optimized shift base matrix having an optimized shift value is generated (707). The optimization iteration 706 continues as long as otherwise specified. The target value is controlled according to the number of iterations (705). For example, the target value is gradually increased to relax the requirement for removing short cycles.
[0069] If the maximum score is greater than the target value, select multiple shift values with scores higher than the other scores, and those shift values are randomly modified by integers evenly distributed from 0 to Q-1 (708). For example, only one worst shift value is selected, or the worst 10% shift values are selected to be modified. After the shift values are modified, the optimization process is repeated (706).
[0070] The hill climbing procedure described above can eliminate short cycles in the protograph polar code so that the message passing algorithm for decoding can be properly performed. For example, the four-stage QC polar code for Q=256 lifting coefficients as shown in Figure 2 can eliminate cycle 6 and cycle 8 loops with the following shifted fundamental matrix optimized by the hill climbing method:
number
[0071] This four-stage QC-polar code with the above fundamental matrix achieves a significant gain of more than 2.4 dB compared to the conventional four-stage polar code, while the computational complexity remains in the same order. Heterogeneous binding
[0072] To further improve the error correction performance, some embodiments use a cyclic redundancy check (CRC) code embedded in the data bits. This modification allows the decoder to easily filter out invalid code words. If at least one of the decoding candidates corresponds to a data sequence with a valid CRC, the most likely candidate with a valid CRC can be selected for estimation. Specifically, the soft-decision decoder output has various confidence levels depending on the magnitude of the leftward LLR message value in the first stage of the decoder. By selecting some worst LLR values of the soft-output decision, we can invert the decision to check the validity of the CRC to improve the error correction ability.
[0073] 8A and 8B show a schematic of embedding at least one CRC code in the data bits supplied to the protograph encoder. For example, FIG. 8A shows an example where a CRC parity sequence is embedded at the end of the information bits to verify the correctness of the decoding. In this example, the bits of a codeword 800 are split into a single data portion 801 containing the actual message data, followed by a single CRC code 802 that is calculated (803) from the data portion 801. For example, a C-bit CRC parity check sequence is appended to a (QK-C)-bit data message. The number of parity bits C is relatively small, resulting in no net code rate loss, i.e. (QK-C) / Q2 n For example, 1-bit parity uses a CRC polynomial of D+1, and 2-bit parity uses D 2 +D+1 CRC polynomial is used, and 3-bit parity is D 3 +D 2 +D+1 CRC polynomial is used, and 4-bit parity is D 4 +D 2 +D+1 CRC polynomial is used, and 8-bit parity is D 8 +D 4 +D 3 +D 2 +D+1 CRC polynomial is used, and 16-bit parity is D 16 +D 2 +1, etc., where D denotes the delay unit. The parity polynomial can be adaptively modified according to the protograph code specification 140.
[0074] FIG. 8B shows another example of using multiple CRC codes, where a first CRC is embedded in the middle of the data bits to improve the performance of polarity decoding, and a second CRC is embedded at the end of the data bits. Such multiple CRC codes embedded in the data bits can be used to verify partial decoding candidates. In this way, the CRC can help the polarity decoder to prune candidate codewords at intermediate steps of the decoding procedure. In addition, the multiple CRC codes can prevent potential error propagation in polarity decoding. The data bits 810 are embedded with multiple CRC codes that divide the data bits into four parts. The first data part 811 is followed by a first CRC part 812 calculated (813) from the first data part 811. The second data part 814 is followed by a second CRC part 815 calculated (816) from the second data part 814.
[0075] Some embodiments split the data bits using more than two CRC codes so that the decoder can check the CRC validity in more segments of the bit sequence. However, using more CRC codes increases the redundant parity bits, which may reduce the net code rate when the codeword length is short. This has been a significant problem for conventional polar codes. The method and system of the present invention is based on the recognition that by using a high lifting factor Q, the rate loss due to multiple CRC codes is significantly reduced. This embodiment also provides the non-obvious advantage that each proto-polarization unit in the first polarization stage can cancel the hard decision for rightward message passing when verifying the CRC across the replicated Q parallel encoders.
[0076]
number
[0077] To obtain more flexibility, some embodiments use a non-uniform combining operation using partial CRC codes. FIG. 8D shows an example embodiment of non-uniform combining. In the embodiment, the Q copies of the polar encoder use non-identical frozen bit positions and code rates. For example, the first Q1 short polar encoder 843 uses R1=K1 / 2 for a first information array of size K1×Q1. n Using a lower code rate, the intermediate Q2 short polarity encoder 844 has a code rate of R2=K2 / 2 for a second data array of size K2×Q2. n The short polarity encoder 845 for the last Q3 uses a medium code rate of R3=K3 / 2 for the last data array of size K3×Q3. n A higher code rate is used. The frozen bit portion is now stepped (842). Rows 846 across the encoder with Q=Q1+Q2+Q3 are encoded with individual CRC codes (847). In some embodiments, the CRC parity bit lengths can be non-uniform and some rows do not have CRC codes embedded. For example, 10% of the columns use a 16-bit CRC code, 20% of the rows use an 8-bit CRC code, and the remaining rows do not use any CRC code. With this uneven combination, u i The reliability per bitplane can be further adjusted as soft-frozen bits. Belief messages can be propagated from the lower-rate encoder to the higher-rate encoder. Some embodiments use a specific scheduling such as windowed decoding from the lower-rate encoder part to the higher-rate encoder part. This windowed decoding can further improve the decoding latency. For some embodiments, in addition to the CRC code, a protograph polarity code is embedded in another error correction code. For example, in such an embodiment, a high-rate BCH code is embedded to remove the error floor.
[0078] FIG. 9 shows a block diagram of protograph polarity decoding with uniform or non-uniform combination when multiple CRC codes are embedded, according to some embodiments. The protograph polarity decoder takes the soft input message 902 and corrects potential errors. The protograph polarity decoder uses Q parallel short polarity decoders 902, where all belief messages are initialized according to the frozen bit positions. After updating one round of message passing across multiple stages and decoders, a hard decision operation 903 is performed on the rightward message in the first polarization stage. The hard decision uses the sign of the belief message to determine a "0" or a "1". In some embodiments, the hard decision is applied probabilistically depending on the magnitude of the belief message. For example, u i The estimate of the i-th data for is determined probabilistically as follows:
number
[0079] This probabilistic hard decision can generate a list of multiple hard decision candidates 903 for the data bit estimates. In some embodiments, a simple bit flip for some low-reliability bits with small magnitude LLR is used to generate the hard decision list. All candidates in the list are verified (904) in each CRC code segment. If all CRC checks pass for validity, the iterative decoding ends (905) and soft output and hard decision results are generated (906). Otherwise, the decoding iterations continue (908) until a predefined maximum count is reached. Some embodiments use incremental freezing bits (907). In this case, some partial hard decision data (or high-reliability belief messages above a threshold) that pass CRC verification are fixed as additional frozen bits (902) to update the next message passing iteration. These incremental freezing bits can accelerate the convergence of the decoding. This is particularly effective when using multiple CRC codes for the protograph polarity code, since partial data can be frozen segmentally. Frozen Bit Design
[0080] Some embodiments are based on the recognition that conventional polar coding structures assume a situation where the communication channel and modulation scheme provide uniform transmission reliability for each transmitted codeword bit. However, some practical systems, such as high-order modulation, frequency-selective fading channels, time-varying channels, and multiple-input multiple-output (MIMO) channels with multiple antennas, result in uneven reliability across transmitted bits. Some embodiments are based on another recognition that polar coding performance is highly dependent on frozen bit positions according to channel statistics. The present invention discloses a method for optimizing frozen bit positions for protograph polar codes in such situations by using protograph-based extrinsic information transfer (P-EXIT) analysis.
[0081] In some embodiments, the transmitter 110 and / or the receiver 130 include a channel estimator configured to determine parameters of the communication channel 120. For example, the parameters of the communication channel include values of uneven reliability for the transmission of bits of the encoded codeword and / or other statistics of the channel, such as SNR and power delay profile. The parameters of the communication channel may be determined using various methods, such as least squares channel estimation or blind power estimation based on pilot and training symbols. The protograph code specification 140 includes a memory for storing a mapping between various values of the code parameters for various values of the parameters of the communication channel. In this manner, the present embodiment may select a combination of values of the code parameters of the protograph polarity code based on the parameters of the communication channel determined by the channel estimator.
[0082] To facilitate protograph decoding, the frozen bit positions are improved so that the polarization effect can be boosted by addressing the statistics of the likelihood during decoding. The frozen bit position design is particularly important for high-order modulation and frequency selective fading channels, where different coded bits are corrupted by different noise strengths, resulting in uneven bit reliability. This embodiment utilizes knowledge of the statistics of the likelihood to select the frozen bit positions to improve the performance of soft decision decoding. In addition, how to map the coded bits to which modulation bits is important for such uneven reliability, because different mappings may reduce the polarization effect. Therefore, in addition to the frozen bit position design, careful interleaving design for mapping the coded bits to the modulation bits is required. The method of the present invention provides a method for jointly designing frozen bit positions and interleaving for such high-order modulation and fading channels.
[0083] FIG. 10A shows an example schematic diagram of a transmitter with interleaving. In this case, the polarity coding 112 of the coded bits (x1, ..., x N ), which are then mapped to different modulation bits by an interleaver 1020 in a QAM modulator, MIMO antenna multiplexing, or OFDM modulator 1010, from the least-significant bit (LSB) plane to the most-significant bit (MSB) plane. The LSB to MSB have different bit reliabilities. In addition to the bit significance, each modulated symbol (t1, ..., t N′ ) may have different channel noise levels, for example due to frequency selective fading channels and multi-antenna transmission. The method of the present invention carefully maps important coded bits to reliable modulation bits so that high decoding performance is achieved.
[0084] Figure 10B shows a procedure for jointly optimizing the interleaver and frozen bit positions for non-uniform channels. The method employs an interleaver scheme in which the permutation performed by the interleaver is determined by a set of parameters that can be optimized to be tractable without considering all possible permutations. For example, one implementation of interleaving is based on polynomial permutation, including a quadratic polynomial permutation (QPP) interleaver, which permutes the coded bit index i (0-based) to the modulation bit index as follows:
number
[0085] First, the interleaver is set to an initial permutation (1001). Then, the polar code structure is optimized for this initial interleaver permutation by selecting the data index set corresponding to the most reliable pseudo channel (1002). Next, the error correction performance of the polar code structure and the interleaver is evaluated (1003). This evaluation can be performed empirically by simulation and / or analytically by error bounds that can be calculated from the reliability of the pseudo channel selected by the data index set. For example, for each polarization operation, the statistics of the likelihood can be tracked by Bhattacharyya parameters, density evolution, Gaussian approximation, or extrinsic information transfer (EXIT) methods. To capture the non-uniform reliability for the protograph polar code, the method of some embodiments uses an extended version of EXIT called P-EXIT with a non-conventional trace.
[0086]
number
[0087] When the mutual information after finite iterative decoding is calculated according to the specified schedule, the error rate at the i-th input bit is
number
[0088] Note that increasing the lifting factor Q makes the P-EXIT analysis more accurate since the variance within the protovariable and protocheck nodes can be inversely reduced. Specifically, the variance of the external information is modeled as
number
[0089] A decision is then made to continue or terminate the iterative optimization procedure based on whether the error correction performance has converged (i.e., does not change significantly with respect to the previous iteration) or whether a limit on the total number of iterations has been reached (1004). If continuing, the interleaver permutation is optimized (1005) while keeping the polar code data set index fixed, then the data set index is optimized again (1002) while keeping the interleaver fixed, then the performance of the protograph polar coding and interleaver is re-evaluated (1003), and a decision is again made to continue or terminate the iterative optimization (1004). After completing these iterations, the final result is a jointly optimized interleaver and polar code structure (1006). This joint optimization of the frozen bit positions and interleaving provides an enhanced polarization effect, especially for wireless fading channels. Irregular pruning
[0090] In the prior art, it was shown that conventional polarity codes can be improved by introducing irregular pruning of polarization units, which can reduce the encoding / decoding complexity, decoding latency, and even error rate due to improved Hamming weight distribution. The method and system of the present invention further improves protograph polarity codes by deactivating polarization units, which provides another advantage with respect to protograph polarity codes. In particular, pruning of proto-polarization units can help remove short cycles.
[0091] FIG. 11A shows an example of an irregular protograph polarity coding structure in which some protopolarization units are inactivated (1110). By carefully selecting the inactive protopolarization units, the error correction performance can be improved and the computational complexity for encoding and decoding can be reduced. In addition, several sets of inactivated protopolarization units allow partially parallel decoding, reducing the latency of decoding. The location of the inactive protopolarization units is determined by analyzing the error bounds with the P-EXIT method, so that the error bounds are minimized in a greedy manner. Since most protopolarizations do not dramatically degrade the decoding performance, this irregular inactivation can significantly reduce the decoding complexity by selecting more inactive protopolarization units.
[0092] In fact, deactivating protopolarization units for protograph polarity codes provides another advantage of removing some short cycles to improve message passing capability. Figure 11B shows an example of an irregular QC polarity coding structure for two-stage polarization to illustrate the advantage provided by deactivating protopolarization units. A regular protograph polarity coding has two protopolarization stages, each stage having two protopolarization units. As shown in Figure 6A, such a protograph can be divided into s 11 →s 22 →s 12 →s 21The irregular QC polarity coding deactivates the first protopolarizer unit in the second stage (1123). This inactive protopolarization can break the cycle 4 loop 1124 regardless of the shift value. The irregular QC polarity coding has a gain of 0.9 dB over the regular polarity code while reducing the complexity by 25%. This example suggests that deactivating the protopolarizer units can reduce the number of short cycles as well as the computational complexity. Note that the short cycles also depend on the frozen bit positions since the cycle length can be eliminated due to infinite confidence in some graph edges, and therefore optimizing the frozen bit positions is as important as optimizing the perimeter.
[0093]
number
[0094] FIG. 12 shows a block diagram of a method for selecting inactive post-polarizers of an irregular QC polarity coding structure according to some embodiments. The method initializes (1201) an activation matrix, called a skeleton base matrix, to "+" (1202) to indicate an active post-polarization unit. Then, the method deactivates (1203) the previous activation matrix, i.e., changes the "+" elements to "-" to indicate deactivation. Deactivation is considered for all possible N' positions 1204. Then, the error rate probability is calculated for each irregular QC polarity coding, where the interleaver and frozen bit positions are optimized in the same way as described in FIG. 11B during the analysis of error rate performance with P-EXIT evolution. The method selects (1206) the best L' irregular QC polarity coding with the smallest error probability. The method further deactivates (1203) various protopolarizer units (1207) for each selected irregular QC polarity coding. This procedure continues (1208) until a termination condition is met. The termination condition includes, for example, when the error rate performance is minimized or when the error rate performance is worse than that of the regular QC polarity coding. After the list tree search is finished, an irregular QC polarity coding with optimized activation table, interleaver, and frozen bit positions is generated (1209). It should be noted that the active protopolarizations are shown as "+" in the skeleton base matrix, and these active shift values are optimized to increase the perimeter in the same manner as described in FIG. 7 after the skeleton base matrix and frozen bit positions are designed. Thus, a set of rules for connecting or disconnecting the multi-stage protopolarization units is defined in the optimized skeleton base matrix.
[0095] For example, for a 6-stage QC polarity code with Q=1024, the error correction performance can be improved over regular replicas by pruning a few tens of percentages of proto-polarization units using the optimization method described in FIG. 12. In addition, the irregular QC polarity code can be superior to conventional polarity codes even when more than 60% of the polarization units are pruned. As a result, the irregular QC polarity code can further reduce the decoding complexity with potential performance improvements. Enhanced Kernels and Replacements
[0096]
number
[0097]
number
[0098] In some embodiments, the protograph polar code is calculated by the lifting operation using a permutation matrix P ij Some embodiments are based on the recognition that the circulant matrix I(s) is not the only permutation matrix of weight 1. For example, another permutation matrix of weight 1 is an inverse circulant matrix that inverts the circulant matrix along the horizontal axis as follows:
number
[0099] In some embodiments, the weight 1 permutation is realized by a polynomial permutation such as a QPP interleaver with various parameters. Even a random permutation is used as a random weight 1 permutation in some embodiments. As another example, in some embodiments,
number
[0100] Some embodiments further increase the freedom in the design space to facilitate soft-decision decoding by spatially combining multiple polar codes in unconventional ways with various parameters such as code length, code rate, and frozen bit positions. Some embodiments use lifting operations to combine replicated encoders in a partially convolutional manner.
[0101] Figure 13 shows an exemplary schematic diagram of spatial coupling. Here, a two-stage polar encoder 1300 is replicated Q times as in normal lifting (1301), but the permutation is applied partially across a subset of the encoders. For example, the first polarization stage is combined for the first two encoders (1302), while the last encoder is left unpermuted (1303). For the second polarization stage, the last two encoders (instead of all Q) are combined (1304). In this way, a spatially coupled polar encoder can be realized.
[0102] FIG. 14 shows another exemplary schematic diagram of spatial combination using a convolutional lifting operation. A two-stage polar encoder 1400 is duplicated (1401). This embodiment arranges the duplicated polar encoders in a staircase fashion. The second polarization stage in the first encoder is combined with the first polarization stage in the second encoder (1402). Similarly, the second polarization stage in the second encoder is combined with the first polarization stage in the third encoder (1403). In this way, encoding and decoding can be performed convolutionally. Specifically, the encoded codeword in the first encoder (1404) depends on the first encoding bits and the second encoding bits, but is independent of the other encoders. This windowed memory feature allows for windowed encoding and decoding. Then, decoding latency can be reduced since an infinitely large number of encoders can be convolutionally combined while encoding and decoding can be performed without waiting to receive the entire codeword. Other examples of spatially coupled protograph polar coding implementations include, but are not limited to, braided structures, cubic convolutions, tail biting, and torus tail biting. In particular, tail biting uses cyclic padding of the input data to duplicate the end of a sequence to the beginning of a sequence, allowing spatial coupling to be encoded without worrying about memory effects in the first convolution.
[0103] In one embodiment, a protograph-based polar coding structure is constructed, and parallel polar codes are mixed between polarizers at different positions by shifting operations. Another embodiment uses an ever-increasing hierarchical structure to provide a rateless function, where only parity bits are continuously generated by the replicated polar codes until the receiver acknowledges the completion of decoding. Thus, the application of an irregular protograph polar coding structure and various component polar codes with different coding parameters provides different degrees of freedom (and thus different degrees of error correction performance) over the entire codeword of the combined code. This overall codeword is serialized for transmission over a communication channel, and this serialization may be replaced by permutations before transmission over a non-uniform channel, possibly to obtain error correction performance gains. Protograph polar type LDGM code
[0104] The method and system of the present invention provides a method for improving various other linear codes by introducing the concept of protographs to lift the generator matrices of base codes. Some embodiments of the present invention further extend the application of the protograph-based design method for GM-based linear codes to realize generalized LDGM codes, which we call polar-type LDGM codes.
[0105] For example, some embodiments improve upon traditional cyclic codes that use a generator polynomial to encode data bits. The well-known Hamming and BCH codes are families of such cyclic codes. In particular, Hamming codes have a cyclic polynomial length of 1+D+D 3 The GM of a Hamming code (systematic) for K=4 information bits and N=7 codeword bits is expressed as follows for a matrix size G of K×N:
number
[0106] In some embodiments, a similar protograph lifting operation is applied to the GM of the above base code with a permutation matrix to increase the codeword by a factor of Q as follows:
number
[0107] This Q-fold lifting can reduce the density of the GM for any linear code, and these are called generalized LDGM codes. If the permutation matrix is cyclic, these are also called QC LDGM codes. Figure 15A shows an example QC LDGM code based on the protograph of the systematic Hamming code described above, with an identity matrix, i.e. P ii The diagonal permutation is chosen so that =I(0). The off-diagonal permutation is decomposed into three polarization units as follows:
number
[0108] The top three bit-planes are frozen to "0" 1551, and the bottom four bit-planes are assigned to information bits 1552. Note that in this figure, for convenience, the bit-planes are in reverse order from u7 to u1. The size of the data array u is increased to N×Q with a lifting factor Q greater than 1. The first polarization stage 1553 is the polarization of the base generator matrix G s After passing through the second stage 1554 and the third stage 1555, an encoded sequence 1556 is generated. Each proto-polarization unit is assigned a cyclic shift value S ij A substitution rule is used with (1557). This protograph has cycle 2, which can be eliminated by shift value optimization. Also note that the protograph in FIG. 15A is not unique from the base generator matrix, since different orders of polarization stages can be constructed. Therefore, to specify the code, a connection rule for constructing the protograph is used.
[0109] For some embodiments, the protograph LDGM code is fully structured with multi-stage polarization. It is known that the GM of a conventional Hamming code can be expressed as follows even for the non-systematic case:
number
[0110] This GM can be further extended to N×N matrices by introducing three zero-freezing bits as a polarity code to encode u=(u1,u2,u3,u4,0,0,0) as follows:
number
[0111]
number
[0112] The above-described embodiments of the present invention have many advantageous features, including but not limited to the following. The encoding and decoding share the same graphical structure specified in the GM without computing the PCM. The encoding and decoding can be fully parallelized. Short cycles in the protograph can be effectively eliminated. Since shorter basic codes are combined, code design can be simplified. Belief propagation decoding can work well in high-period designs. The computational complexity for the protograph decoding is kept as low as that of the original short code. State-of-the-art performance is achievable. Any linear code can be used as a base protograph and converted into a polar-type LDGM code with multiple polarization stages. SC and BP decoding can be combined with incremental freezing bits if a partial CRC code is embedded across the replicated encoder. Irregular coding is easy to employ with protopolarization pruning. Non-binary coding can be easily applied. Spatial combining can be used to adjust the decoding parallelism and the decoding latency. Established techniques used for protograph LDPC codes can be used for protograph LDGM codes, such as the P-EXIT method and the perimeter maximization method, which are established techniques used for rotograph LDPC codes.
[0113] Most of the above advantages are not available for conventional polar codes and conventional protograph LDPC codes. In particular, the highly parallel encoding and decoding capabilities are a great advantage of the present invention for practical applications requiring high throughput processing, such as optical fiber communications, which require data rates exceeding terabits per second.
[0114] FIG. 16A shows an example of a highly parallel and pipelined implementation of a polar-type LDGM encoder 1600 for high-throughput transmission according to the present invention. For example, a QC polar code operates n=8 stages of Q=256 polar encoders 1602 in parallel, and then N=65536 bits are generated as encoded codewords 1603 given source data 1601 of size K×Q. Note that the lifting permutation to combine multiple encoders does not involve any computational complexity, but does involve data exchange between encoders at each stage. Since each elementary encoder 1602 has n=8 shallow stages, encoding can be achieved in 8 clocks to encode through 8 stages. In this way, highly parallelizing the encoding for all N bits generates large codewords with a small number of 8 stages of pipelining. Therefore, the highly parallel implementation of the above-mentioned polarity-based LDGM encoder can improve the efficiency of the encoding process of the source data and can also improve the performance of the processor (hardware processor) in the transmitter, thereby increasing the throughput of the transmitter.
[0115] FIG. 16B shows an example of a highly parallel pipelined implementation of a polarity-type LDGM decoder for high throughput processing, according to an embodiment. For example, the polarity-type LDGM decoder 1610 is fully parallelized for Q=256 short decoders 1612, regardless of the message passing schedule (including SCL decoding or BP decoding). Each decoder takes a soft input belief message 1611 and generates a soft output message 1614 by message passing over multiple stages via a permutation interconnect 1613 to exchange data across the parallel decoders. Some embodiments further use pipelining the same decoding process with additional decoders 1620 to increase the decoding throughput, rather than repeating the message passing in the same decoder 1610. For example, the first decoding iteration is performed in the first set of parallel decoders 1610, and the second decoding iteration is performed in the second set of Q parallel decoders 1620. Further iterations are spread out to the pipelined decoder via a serialized interconnect 1623. It should be noted that each decoder 1612, 1622 can be processed in parallel since belief messages are simply exchanged between the various decoders without extra complexity. The ability of parallel and pipeline implementation is a prominent advantage of polarity-type LDGM codes for high throughput transceivers. Thus, the implementation of the polarity-type LDGM decoder described above can improve the effectiveness of the codeword decoding process and can increase the performance of the processor (hardware processor) in the receiver, thereby increasing the throughput of the receiver. Other relevant features
[0116] 17A illustrates an example block diagram of a system suitable for implementing various components of a receiver for performing soft decoding according to some embodiments and / or a transmitter for encoding codewords according to some embodiments. The system 1700 may include one or a combination of a sensor 1710, an inertial measurement unit (IMU) 1730, a processor 1750, a memory 1760, a transceiver 1770, and a display / screen 1780 that may be operably coupled to other components via a connection 1720. The connection 1720 may include a bus, a line, a fiber, a link, or a combination thereof.
[0117] The transceiver 1770 may include, for example, a transmitter capable of transmitting one or more signals over one or more types of communication networks and a receiver for receiving one or more signals transmitted over one or more types of communication networks. The transceiver 1770 may enable communication with wired or wireless networks based on various technologies, such as, but not limited to, femtocells, Wi-Fi networks or wireless local area networks (WLANs) that may be based on the IEEE 802.11 family of standards, wireless personal area networks (WPANS) such as Bluetooth, near field communication (NFC), networks based on the IEEE 802.11 family of standards, and / or wireless wide area networks (WWANs) such as LTE, WiMAX, etc. The system 1700 may also include one or more ports for communicating over a wired network, such as optical fiber communication.
[0118] In some embodiments, the processor 1750 can also receive input from the IMU 1730. In other embodiments, the IMU 1730 can include a three-axis accelerometer, a three-axis gyroscope, and / or a magnetometer. The IMU 1730 can provide velocity, orientation, and / or other position-related information to the processor 1750. In some embodiments, the IMU 1730 can output measured information synchronized with the capture of each image frame by the sensor 1710. In some embodiments, the output of the IMU 1730 is used in part by the processor 1750 to fuse the sensor measurements and / or to further process the fused measurements.
[0119] The system 1700 may also include a screen or display 1780 for rendering images, such as color and / or depth images. In some embodiments, the display 1780 may be used to display live images captured by the sensor 1710, fused images, augmented reality (AR) images, graphical user interfaces (GUIs), and other program output. In some embodiments, the display 1780 may include and / or house a touch screen that allows a user to input data through some combination of a virtual keyboard, icons, menus, or other GUIs, user gestures, and / or input devices, such as a stylus and other writing means. In some embodiments, the display 1780 may be implemented using a liquid crystal display or a light emitting diode (LED) display, such as an organic LED (OLED) display. In other embodiments, the display 1780 may be a wearable display.
[0120] The example system 1700 can also be modified in various ways in accordance with this disclosure, such as by adding, combining, or omitting one or more of the illustrated functional blocks. For example, in some configurations, the system 1700 does not include the IMU 1730 or the sensors 1710. In some embodiments, portions of the system 1700 take the form of one or more chipsets, etc. and / or the like.
[0121] The processor 1750 may be implemented using a combination of hardware, firmware, and software. The processor 1750 may represent one or more circuits configurable to execute at least a portion of the computational procedures or processes associated with sensor fusion and / or methods for further processing the fused measurements. The processor 1750 retrieves instructions and / or data from the memory 1760. The processor 1750 may be implemented using one or more application specific integrated circuits (ASICs), central processing units (CPUs), graphic processing units (GPUs), digital signal processors (DSPs), DSP devices, programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, embedded processor cores, electronic devices, other electronic units designed to perform the functions described herein, or combinations thereof.
[0122] The memory 1760 may be implemented within the processor 1750 and / or external to the processor 1750. As used herein, the term "memory" refers to any type of long-term, short-term, volatile, non-volatile, or other memory, and is not limited to any particular type or number of memories or the type of physical medium on which the memories are stored. In some embodiments, the memory 1760 holds program code that facilitates soft decoding and polarity coding.
[0123] In some embodiments, in addition to or instead of soft decoding, the processor 1750 can execute one or a combination of soft decoding applications 1755. For example, the soft output of the decoding can be used to decode a concatenated ECC formed from multiple component ECCs combined into a higher performance code. Another example is a system using iterative equalization and decoding, where the soft decision output from the decoder is sent back to the demodulator to iteratively refine the decoder input. Yet another example is to act on the decoded output, such as showing the output on the display 1780, storing the output in memory 1760, transmitting the output using the transceiver 1770, and / or performing an action based on the output and measurements of the sensor 1710.
[0124] Further, another embodiment of the present invention can provide an encoder circuit for generating encoded digital data from input digital data. FIG. 17B is a schematic diagram showing a block diagram of an encoder circuit 112 for generating encoded digital data from input digital data. The encoder circuit 112 may also be referred to as an encoder 112. The encoder circuit 112 is configured to generate encoded digital data from a source bitstream as input digital data. The encoder circuit 112 may include an encoder interface 1120 configured to receive a source bitstream 111, at least one memory bank 1260 configured to store a computer-executable encoding method 1255 (program / computer-executable instructions / or equivalent circuit), and one or more computing processors 1250 configured to generate codewords from the input digital data as encoded digital data by performing steps of the computer-executable encoding method. In this case, the computer-executable encoding method 1255 is configured to cause the processor 1250 to perform steps including a step of defining a protograph structure of a polar-type low-density generator matrix (LDGM) code based on a code specification. The code specification includes at least two stages of proto-polarization, at least one rule of frozen bit allocation, at least one rule of protograph substitution, and at least one rule of message passing. The method steps of the computer executable encoding method 1255 further include initializing a data array with a source bit stream and a set of frozen bits according to the at least one rule of frozen bit allocation, and propagating the data array according to the at least one rule of message passing throughout the stages of proto-polarization.The steps further include (a) sequentially feeding the data sequence to at least two stages of proto-polarization from one stage to another, each stage of proto-polarization including at least one proto-polarization unit, (b) permuting the data sequence in each of the proto-polarization units according to at least one rule of protograph permutation, and (c) generating the codeword as encoded digital data based on a specified order by modifying the data sequence in each of the proto-polarization units according to at least one rule of message passing and arranging the modified data sequence. In some embodiments, the generated digital data is modulated by the modulator 113 and transmitted from the communication channel 120 via the front-end circuit 114.
[0125] 17C is a schematic diagram illustrating a block diagram of a decoder circuit (or simply, decoder) 133 for decoding encoded digital data according to some embodiments of the present invention. The decoder circuit 133 is configured to generate decoded digital data from the encoded digital data represented by the encoded (noisy) codeword received from the communication channel 120.
[0126] The decoder circuit 133 includes a decoder interface 1320 configured to receive as input data an encoded codeword from a demodulator 132 for demodulating a noisy version of the modulated encoded digital data, at least one memory bank 1360 configured to store a computer executable decoding method (program / computer executable instructions / or equivalent circuit) 1355, and one or more computing processors 1350. In this case, the computer executable decoding method (program) 1355 causes the one or more computing processors 1350 to generate decoded digital data from the encoded codeword from the decoder interface 1320 by executing steps of the computer executable decoding method. According to the computer executable decoding method, the one or more computing processors 1350 execute steps including a step of defining a protograph structure of a polar-type low-density generator matrix (LDGM) code based on a code specification. The code specification includes at least two stages of proto-polarization, at least one rule of frozen bit allocation, at least one rule of protograph substitution, and at least one rule of message passing. The steps further include initializing an array of leftward and rightward messages associated with each edge in the protograph structure, where the rightward messages to be fed to a first stage of proto-polarization are fixed based on the at least one rule of frozen bit allocation, and the leftward messages to be fed to a last stage of proto-polarization are fixed in a specified order based on the encoded codeword. The steps further include propagating the leftward and rightward messages according to the at least one rule of message passing across at least two stages of proto-polarization including at least one variable node and at least one check node. The leftward and rightward messages are updated by an iterative method.The iterative method includes the steps of (a) selecting a parallelizable set of variable nodes and check nodes according to at least one rule of message passing, (b) feeding a set of leftward and rightward messages as incoming messages to the selected variable nodes and check nodes according to at least one rule of protograph substitution, (c) computing a set of outgoing messages based on the incoming messages at the selected variable nodes and check nodes according to at least one rule of message passing, (d) replacing the set of leftward and rightward messages with the set of outgoing messages at the selected variable nodes and check nodes, and (e) repeating steps (a)-(d) a specified number of times according to at least one rule of message passing. Further, the one or more processors 1350 generate an output message as decoded digital data by selecting the set of leftward messages in a specified order according to a computer-executable decoding method 1355.
[0127] All of the above-described embodiments of the present invention can be implemented in any of numerous ways. For example, the embodiments can be implemented using hardware, firmware, software, or a combination thereof. If implemented in software, the software code can be executed on any suitable processor or collection of processors, whether provided on a single computer or distributed among multiple computers. Such a processor can be implemented as an integrated circuit with one or more processors in an integrated circuit component. However, the processor can be implemented using circuitry in any suitable format.
[0128] Also, embodiments of the invention may be implemented as a method, an example of which has been provided above. The operations performed as part of the method may be ordered in any suitable manner. Thus, embodiments may be constructed in which operations are performed in an order different from that illustrated, which may include performing some operations simultaneously although shown as sequential operations in the exemplary embodiment.
[0129] The use of ordinal terms such as "first," "second," etc. in the claims to modify claim elements does not, in itself, imply a priority, precedence, or order of one claim element over another, nor does it imply a chronological order of acts in a method of performing, but is merely used as a label to distinguish one claim element having a particular name from another element having the same name (but using ordinal terms) to distinguish among the claim elements.
[0130] Although the invention has been described by way of examples of preferred embodiments, it is to be understood that various other adaptations and modifications can be made within the spirit and scope of the invention.
[0131] It is therefore intended in the appended claims to cover all such changes and modifications that fall within the true spirit and scope of this invention.
Claims
1. 1. A computer-implemented method for encoding digital data, executed by one or more computing processors, comprising: defining a protograph structure of a polar-type low-density generator matrix (LDGM) code based on a code specification, the code specification including at least two stages of proto-polarization, at least one rule of frozen bit allocation, at least one rule of protograph substitution, and at least one rule of message passing, the computer-implemented method further comprising: accessing a source bitstream as input digital data; initializing a data array with the source bitstream and a set of frozen bits according to the rules of the frozen bit allocation; and propagating the data array according to the message passing rules throughout the stages of the proto-polarization, the propagating step further comprising: (a) sequentially feeding said data array to a proto-polarisation stage, said proto-polarisation stage comprising at least one proto-polarisation unit, (b) permuting said data sequence in each of said proto-polarized units according to said protograph permutation rules; (c) modifying the data array in each of the proto-polarization units according to the message passing rules, the computer-implemented method further comprising: placing the modified data array into a codeword in a specified order; providing said codeword as an encoded digital data representation of said input digital data.
2. The proto-polarizing unit is At least one protocheck node; at least one proto variable node; The method of claim 1 , wherein at least one pair of the proto check node and the proto variable node are interconnected to pass multi-bit messages according to rules of the protograph substitution.
3. 2. The method of claim 1, wherein modifying the data array based on the message passing rules comprises applying a set of algebraic arithmetic operations at each of the proto-polarized units, the algebraic arithmetic operations being based on a finite ring, a finite Galois field, a look-up table, or a combination thereof.
4. 2. The method of claim 1, wherein the code specification defining the protograph structure of the polar-type LDGM code is based on a lifting operation on a generator matrix of a base code, the lifting operation applying duplication and permutation of the base code by replacing each element of the generator matrix with a permutation matrix, and the generator matrix is decomposed into at least two sub-generator matrices to define multiple proto-polarization stages.
5. 2. The method of claim 1, wherein the protograph permutation rules include a set of parameters that define a cyclic permutation with a shift value, an inverse cyclic permutation with a shift value, a polynomial permutation with a set of polynomial coefficients, a permutation matrix of random weight 1, a permutation matrix of random weight 2, a permutation matrix of random weight 3, or a combination thereof, associated with each of the proto-polarizations.
6. The rule of frozen bit allocation and the rule of protograph substitution are determined by an iterative method, the iterative method comprising: (a) initializing a set of frozen bit positions, a skeleton base matrix, and permutation parameters; (b) pruning the set of proto-polarization units according to the skeleton elementary matrix; (c) calculating an upper bound on the error rate based on a protograph extrinsic information transfer (P-EXIT) method, the P-EXIT method further comprising: (c1) initializing a set of extrinsic information associated with each edge in a protograph as a function of said frozen bit positions and said replacement parameters given a set of channel mutual information; (c2) updating the extrinsic information in the proto-polarization units based on a set of P-EXIT rules that define the evolution of the extrinsic information; (c3) repeating said updating step (c2) for said set of proto-polarization units in a specified order according to a decoding schedule; (c4) calculating the upper bound on an error rate based on the updated extrinsic information and the frozen bit positions, the iterative method further comprising: (d) calculating a score for each of the protopolarization units, said score comprising: (dl) finding cycles leading away from a protopolarized unit and returning the same protopolarized unit through said protograph; (d2) determining a cycle length according to the frozen bit positions, the skeleton base matrix, and the permutation parameters; (d3) adding a weighting factor according to the cycle length to all of the scores associated with the protopolarization units involved in the cycle; (d4) repeating steps (d1)-(d3) for a designated set of paths in the protograph; and and the iterative method further comprises: (e) modifying the frozen bit positions, the skeleton base matrix, and the replacement parameters based on the upper bound and the score; and (f) repeating steps (b) through (e) until specified conditions regarding said upper limit and said score are met.
7. The step of initializing the data array based on the rules of the frozen bit allocation further comprises: (a) forming a selectively selected bit sequence from a portion of the data array; (b) calculating a parity sequence from the bit sequence based on a linear code, including a cyclic redundancy check (CRC) code, a Bose-Chaudhuri-Hocquenghem (BCH) code, a cyclic code, a quasi-cyclic code, a convolutional code, or a combination thereof; (c) embedding the parity sequence into a selected portion of the data array; and (d) repeating steps (a)-(c) to embed a plurality of parity bits into said data array a specified number of times.
8. 5. The method of claim 4, wherein the base code is a polar code including at least two polarization stages based on at least two Kronecker powers of a kernel matrix, the kernel matrix being a full rank matrix of order size greater than one, and the kernel matrix being based on a finite Galois field, a finite ring, or a combination thereof.
9. 2. The method of claim 1, wherein the protograph structure defined by the code specification is based on spatial coupling of the proto-polarized units, the spatial coupling forming the protograph structure uniformly or non-uniformly to construct a rectangular, cubic, braided, staircase, or torus shape of the protograph, with or without tail biting.
10. 1. A computer-implemented method for decoding noisy codewords, executed by one or more computing processors, comprising: defining a protograph structure of a polar-type low-density generator matrix (LDGM) code based on a code specification, the code specification including at least two stages of proto-polarization, at least one rule of frozen bit allocation, at least one rule of protograph substitution, and at least one rule of message passing, the computer-implemented method further comprising: accessing input data representing belief messages about noisy codewords; and initializing an array of leftward and rightward messages associated with each edge in the protograph, wherein the rightward messages provided to a first stage of proto-polarization are fixed based on the frozen bit allocation rules and the leftward messages provided to a last stage of proto-polarization are fixed based on the input data in a specified order, the computer-implemented method further comprising: propagating the leftward message and the rightward message according to a message passing rule over a plurality of stages of the proto-polarization including at least one variable node and at least one check node, the leftward message and the rightward message being updated by an iterative method comprising: (a) selecting a parallelizable set of variable nodes and check nodes according to the message passing rules; (b) supplying the set of leftward messages and the set of rightward messages as incoming messages to the selected variable nodes and check nodes according to rules of the protograph replacement; (c) computing a set of outgoing messages based on the incoming messages at the selected variable nodes and check nodes according to the message passing rules; (d) replacing the set of leftward messages and the set of rightward messages with the outgoing messages at the selected variable nodes and check nodes; (e) repeating steps (a) through (d) a specified number of times in accordance with the message passing rules, the computer-implemented method further comprising: selecting the set of leftward messages as output messages in a specified order; and providing the output message as decoded digital data.
11. The proto-polarizing unit is At least one protocheck node; at least one proto variable node; The method of claim 10 , wherein at least one pair of the proto check node and the proto variable node are interconnected to pass multi-bit messages according to rules of the protograph substitution.
12. 11. The method of claim 10, wherein computing the outgoing messages at variable nodes and check nodes is based on the message passing rules, rules defining a variable node decoder, and rules defining a check node decoder that are based on a sum-product algorithm, a delta-min algorithm, a min-sum algorithm, an offset min-sum algorithm, a normalized min-sum algorithm, a modified min-sum algorithm, damped min-sum, expanded min-sum, a quantized lookup table, or variations thereof.
13. 11. The method of claim 10, wherein the code specification defining the protograph structure of the polar-type LDGM code is based on a lifting operation of a generator matrix of a base code, the lifting operation applying duplication and permutation of the base code by replacing each element of the generator matrix with a permutation matrix, and the generator matrix is decomposed into at least two sub-generator matrices to define multiple proto-polarization stages.
14. 11. The method of claim 10, wherein the protograph permutation rules include a set of parameters that define a cyclic permutation with a shift value, an inverse cyclic permutation with a shift value, a polynomial permutation with a set of polynomial coefficients, a permutation matrix of random weight 1, a permutation matrix of random weight 2, a permutation matrix of random weight 3, or a combination thereof, associated with each of the proto-polarizations.
15. The step of propagating the leftward message and the rightward message further comprises: (a) generating a finite list of hard-decision codewords based on the leftward messages arriving from a first stage of proto-polarization, either deterministically or probabilistically; (b) forming a selectively selected bit sequence from a portion of the hard-decision codeword list; (c) calculating the reliability of the bit sequence based on a linear code, including a Cyclic Redundancy Check (CRC) code, a Bose-Chaudhuri-Hocquenghem (BCH) code, a cyclic code, a quasi-cyclic code, a convolutional code, or a combination thereof; (d) repeating steps (b) and (c) a specified number of times; and (e) determining a most likely sequence of the hard-decision codewords based on the calculated reliabilities; and and (f) updating the portion of the rightward message provided to the first stage of the proto-polarization based on the most likely sequence of the hard-decision codewords as a function of the reliability.
16. 14. The method of claim 13, wherein the base code is a polar code including at least two polarization stages based on at least two Kronecker powers of a kernel matrix, the kernel matrix being a full rank matrix of order size greater than one, and the kernel matrix being based on a finite Galois field, a finite ring, or a combination thereof.
17. 11. The method of claim 10, wherein the protograph structure defined by the code specification is based on spatial coupling of the proto-polarized units, the spatial coupling forming the protograph structure uniformly or non-uniformly to construct a rectangular, cubic, braided, staircase, or torus shape of the protograph, with or without tail biting.
18. 11. The method of claim 10, wherein the message passing rules further comprise a decoding schedule based on a round robin schedule, a flooding schedule, an adaptive schedule, a windowed decoding schedule, a randomized schedule, a serialized schedule, a successive elimination schedule, or a combination thereof.
19. 1. A system for transferring digital data over a communications channel, comprising: at least one computing processor; at least one memory bank; a transmitter, the transmitter comprising: a source that accepts source data to be transferred; and an encoder for encoding the source data, the encoder generating encoded codewords by performing the method of claim 1, the transmitter further comprising: a modulator for converting the encoded codeword into a modulated signal, the modulated signal being based on amplitude modulation, phase modulation, quadrature amplitude modulation, phase shift keying, frequency modulation, orthogonal frequency division multiplexing, space division multiplexing, pulse position modulation, amplitude shift keying, frequency shift keying, or pulse amplitude modulation, the transmitter further comprising: a front-end interface for transmitting the modulated signal to the communication channel, the system further comprising: a communications channel, including a wired media channel, a wireless media channel, a fiber optic channel, a vacuum free space channel, or a data storage medium; a receiver, the receiver comprising: a front-end interface for receiving an output signal from the communication channel; a demodulator for demodulating the output signal of the channel, the demodulator generating likelihood information as an input belief message, the receiver further comprising: a decoder for decoding the input belief message, the decoder generating a decoded codeword by performing the method of claim 10, the receiver further comprising: a sink for providing the decoded codeword, the system further comprising: a controller for defining code specifications for a polar-type low-density generator matrix (LDGM) code to be used in the encoder and decoder in response to knowledge of the channel, the code specifications including at least two stages of proto-polarization, at least one rule of frozen bit allocation, at least one rule of protograph substitution, and at least one rule of message passing for encoding and decoding.
20. 20. The system of claim 19, wherein the decoder further comprises an unfolded set of pipelined decoders with a set of serialized interconnects, each of the pipelined decoders further comprising a set of parallel decoders with a set of permutation interconnects.
21. 20. The system of claim 19, wherein the encoder further comprises an unfolded set of pipelined encoders with a set of serialized interconnects across multiple stages of proto-polarization, each of the pipelined encoders further comprising a set of parallel encoders with a set of permutation interconnects.
22. 1. An encoder circuit configured to generate encoded digital data from a source bitstream as input digital data, the encoder circuit comprising: an encoder interface configured to receive the source bitstream; at least one memory bank configured to store a computer executable encoding method; and one or more computing processors configured to generate codewords as encoded digital data from the input digital data by performing steps of the computer-executable encoding method, the steps comprising: The method includes the step of defining a protograph structure of a polar-type low-density generator matrix (LDGM) code based on a code specification, the code specification including at least two stages of proto-polarization, at least one rule of frozen bit allocation, at least one rule of protograph substitution, and at least one rule of message passing, the step further comprising: initializing a data array with the source bitstream and a set of frozen bits according to at least one rule of the frozen bit allocation; and propagating the data array according to at least one rule of message passing throughout the proto-polarization stages, the propagating step further comprising: (a) sequentially supplying said data array to at least two stages of said protopolarisation, each stage of said protopolarisation comprising at least one protopolarisation unit, and (b) permuting said data sequence in each of said proto-polarized units according to at least one rule of said proto-polarized units; (c) modifying the data array in each of the proto-polarization units according to at least one rule of the message passing, said step further comprising: an encoder circuit that includes the step of arranging the modified data array to generate the codeword as the encoded digital data based on a specified order.
23. 23. The encoder circuit of claim 22, wherein the encoder interface is configured to transmit the codeword to a modulator circuit, the modulator circuit modulating the codeword for transmission over a communication channel as the encoded digital data.
24. 23. The encoder circuit of claim 22, wherein performing the steps of the computer-executable encoding method is parallelized and serialized for a set of encoding methods by data exchange according to the rules of protograph substitution.
25. 1. A decoder circuit configured to generate decoded digital data from a noisy version of an encoded codeword, the decoder circuit comprising: a decoder interface configured to receive the noisy encoded codeword as input data; at least one memory bank configured to store a computer executable decoding method; and one or more computing processors configured to generate the decoded digital data from the encoded codewords from the decoder interface by performing steps of the computer-executable decoding method, the steps comprising: The method includes the step of defining a protograph structure of a polar-type low-density generator matrix (LDGM) code based on a code specification, the code specification including at least two stages of proto-polarization, at least one rule of frozen bit allocation, at least one rule of protograph substitution, and at least one rule of message passing, the step further comprising: initializing an array of leftward and rightward messages associated with each edge in the protograph structure, the rightward messages provided to a first stage of the proto-polarization being fixed based on at least one rule of the frozen bit allocation and the leftward messages provided to a last stage of the proto-polarization being fixed in a specified order based on the encoded codeword, the step further comprising: propagating the leftward and rightward messages according to at least one rule of message passing over at least two stages of the proto-polarization including at least one variable node and at least one check node, the leftward and rightward messages being updated by an iterative method comprising: (a) selecting a parallelizable set of variable nodes and check nodes according to at least one rule of message passing; (b) supplying the set of leftward messages and the set of rightward messages as incoming messages to the selected variable nodes and check nodes according to at least one rule of the protograph substitution; (c) computing a set of outgoing messages based on the incoming messages at the selected variable nodes and check nodes according to at least one rule of message passing; (d) replacing the set of leftbound messages and the set of rightbound messages with the set of outgoing messages at the selected variable nodes and check nodes; (e) repeating steps (a) through (d) a specified number of times in accordance with at least one rule of the message passing; and generating an output message as decoded digital data by selecting the set of leftward messages in a specified order.
26. 26. The decoder circuit of claim 25, wherein the decoder interface is connected to a demodulator circuit that demodulates the encoded codewords from a communication channel to provide initial belief messages for decoding, the initial belief messages being based on log-likelihood ratios, log-likelihoods, likelihoods, soft decision values, hard decision values, or variations thereof.
27. 26. The decoder circuit of claim 25, wherein the step of executing the computer-executable decoding method steps is expanded into pipelined steps to execute the decoding method with a set of serialized interconnections over multiple iterations, and each of the pipelined decoding methods is further parallelized into a set of decoding methods with a set of permutation interconnections to increase decoder throughput.
Citation Information
Patent Citations
Transmitter, method and non-transitory computer-readable storage medium for transmitting encoded codewords - Patents.com
JP2020501472A
Soft-Output Decoding of Codewords Encoded with Polar Code
US20180226999A1
