A communication device that applies an error in an upper layer in addition to error correction in a physical layer is provided. The communication device includes an acquisition unit that acquires control information regarding forward error correction (FEC) of an upper layer and control information regarding FEC of a lower layer, an encoding-decoding unit that performs error correction encoding or decoding of an information sequence in the upper layer according to control information regarding the FEC of the upper layer, and a puncturing processing unit that performs puncturing or depuncturing in the upper layer. The information sequence after FEC encoding of the upper layer is divided into blocks, and puncturing and interleaving are performed in units of blocks.

Methods and apparatus for training a Neural Network to recover a codeword of a Forward Error Correction (FEC) code are provided. Trainable parameters of the Neural Network are optimised to minimise a loss function. The loss function is calculated by representing an estimated value of the message bit output from the Neural Network as a probability of the value of the bit in a predetermined real number domain and multiplying the representation of the estimated value of the message bit by a representation of a target value of the message bit. Training a neural network may be implemented via a loss function.

A quantum error correcting code with dynamically generated logical qubits is provided. When viewed as a subsystem code, the code has no logical qubits. Nevertheless, the measurement patterns generate logical qubits, allowing the code to act as a fault-tolerant quantum memory. Each measurement can be a two-qubit Pauli measurement.

A transmitter is provided. The transmitter includes: an outer encoder configured to encode input bits to generate outer-encoded bits including the input bits and parity bits; a zero padder configured to constitute Low Density Parity Check (LDPC) information bits including the outer-encoded bits and zero bits; and an LDPC encoder configured to encode the LDPC information bits, wherein the LDPC information bits are divided into a plurality of bit groups, and wherein the zero padder pads zero bits to at least some of the plurality of bit groups, each of which is formed of a same number of bits, to constitute the LDPC information bits based on a predetermined shortening pattern which provides that the some of the plurality of bit groups are not sequentially disposed in the LDPC information bits.

Methods and apparatus for training a neural network to recover a codeword and for decoding a received signal using a neural network are disclosed. According to examples of the disclosed methods, a syndrome check is introduced at even layers of the neural network during the training, testing and online phases. During training, optimisation of trainable parameters of the neural network is ceased after optimisation at the layer at which the syndrome check is satisfied. Examples of the method for training a neural network may be implemented via a proposed loss function. During testing and online phases, propagation through the neural network is ceased at the layer at which the syndrome check is satisfied.

A system for software error-correcting code (ECC) protection or compression of original data using ECC data in a first memory is provided. The system includes a processing core for executing computer instructions and accessing data from a main memory, and a non-volatile storage medium for storing the computer instructions. The software ECC protection or compression includes: a data matrix for holding the original data in the first memory; a check matrix for holding the ECC data in the first memory; an encoding matrix for holding first factors in the main memory, the first factors being for encoding the original data into the ECC data; and a thread for executing on the processing core. The thread includes a Galois Field multiplier for multiplying entries of the data matrix by an entry of the encoding matrix, and a sequencer for ordering operations using the Galois Field multiplier to generate the ECC data.

A transmitter is provided. The transmitter includes: an outer encoder configured to encode input bits to generate outer-encoded bits including the input bits and parity bits; a zero padder configured to constitute Low Density Parity Check (LDPC) information bits including the outer-encoded bits and zero bits; and an LDPC encoder configured to encode the LDPC information bits, wherein the LDPC information bits are divided into a plurality of bit groups, and wherein the zero padder pads zero bits to at least some of the plurality of bit groups, each of which is formed of a same number of bits, to constitute the LDPC information bits based on a predetermined shortening pattern which provides that the some of the plurality of bit groups are not sequentially disposed in the LDPC information bits.

An operation method of a receiving node may include performing a decoding operation for calculating first and second output transform values corresponding to first and second unit output nodes in each of a plurality of operation units constituting the polar decoder, based on first and second input transform values corresponding to first and second unit input nodes, and the decoding operation may include setting initial values of first and second variables for calculating the first output transform value; performing an iterative loop operation for updating the first and second variables; and calculating the first output transform value based on values of the first and second variables updated until a time when the iterative loop operation is terminated, wherein the iterative loop operation is terminated without performing iterations in which the first and second variables are determined not to be updated among a plurality of iterations.

A soft input decoding method and a decoder for generalized concatenated (GC) codes. The GC codes are constructed from inner nested block codes, such as binary Bose-Chaudhuri-Hocquenghem, BCH, codes and outer codes, such as Reed-Solomon, RS, codes. In order to enable soft input decoding for the inner block codes, a sequential stack decoding algorithm is used. Ordinary stack decoding of binary block codes requires the complete trellis of the code. In one aspect, the present invention applies instead a representation of the block codes based on the trellises of supercodes in order to reduce the memory requirements for the representation of the inner codes. This enables an efficient hardware implementation. In another aspect, there is provided a soft input decoding method and device employing a sequential stack decoding algorithm in combination with list-of-two decoding which is particularly well suited for applications that require very low residual error rates.

An accelerated erasure coding system includes a processing core for executing computer instructions and accessing data from a main memory, and a non-volatile storage medium for storing the computer instructions. The processing core, storage medium, and computer instructions are configured to implement an erasure coding system, which includes: a data matrix for holding original data in the main memory; a check matrix for holding check data in the main memory; an encoding matrix for holding first factors in the main memory, the first factors being for encoding the original data into the check data; and a thread for executing on the processing core. The thread includes: a parallel multiplier for concurrently multiplying multiple entries of the data matrix by a single entry of the encoding matrix; and a first sequencer for ordering operations through the data matrix and the encoding matrix using the parallel multiplier to generate the check data.