A bit interleaver, a bit-interleaved coded modulation (BICM) device and a bit interleaving method are disclosed herein. The bit interleaver includes a first memory, a processor, and a second memory. The first memory stores a low-density parity check (LDPC) codeword having a length of 64800 and a code rate of 2/15. The processor generates an interleaved codeword by interleaving the LDPC codeword on a bit group basis. The size of the bit group corresponds to a parallel factor of the LDPC codeword. The second memory provides the interleaved codeword to a modulator for 1024-symbol mapping.

A storage device includes a recording medium, a first memory storing first data read from the recording medium, and a controller. The controller searches for read target data in the first data by executing a parity check on second data that is in the first data and starts at a first position, while executing the parity check, determining whether or not an interruption condition is satisfied, storing the second data in a second memory when the parity check completes without the interruption condition being satisfied and a result of a completed parity check satisfies a first condition, and executing a parity check on third data that is in the first data and starts at a second position, responsive to the interruption condition being satisfied and responsive to the result of the completed parity check not satisfying the first condition.

A data storage device is disclosed wherein a first plurality of codewords are generated each comprising a plurality of symbols, and a first parity sector is generated over the first plurality of codewords. A second plurality of codewords are generated each comprising a plurality of symbols, and a second parity sector is generated over the second plurality of codewords. A third parity sector is generated over a first subset of the first plurality of codewords and a first subset of the second plurality of codewords, and a fourth parity sector is generated over a second subset of the first plurality of codewords and a second subset of the second plurality of codewords. When processing of a first codeword fails, the first codeword and the first parity sector are processed using a LDPC type decoder, and the first codeword and the third parity sector are processed using the LDPC type decoder.

Systems and methods are disclosed for categorizing error types encountered in data access operations based on bit information from a data segment. An example apparatus includes a circuit configured to perform error recovery for one or more data segments including determining an error recovery operation of a plurality of error recovery operations to perform based on bit information of the one or more data segments. The example circuit also performs the determined error recovery operation.

A data storage device is disclosed comprising a head actuated over a disk comprising a data track having at least a first data segment and a second data segment. A first plurality of codewords are generated, and a first parity sector is generated over the first plurality of codewords. The first plurality of codewords and the first parity sector are written to the first data segment. A second plurality of codewords are generated, and a second parity sector is generated over the second plurality of codewords. The second plurality of codewords and the second parity sector are written to the second data segment. During a read operation the data segments of the data track are processed sequentially to decode the codewords using a low density parity check (LDPC) decoder, wherein the reliability metrics of un-converged codewords are stored in a codeword buffer and updated using the respective parity sector.

Technology for improved data detection using machine learning may include a method in which an analog read signal comprising data read from a non-transitory storage medium of the data storage device is received. The analog read signal is processed into a plurality of digital samples. A digital sample from the plurality of digital samples is classified into a category from a plurality of categories using a machine learning algorithm for at least some of the plurality of digital samples. The plurality of digital samples is then decoded based on at least some of the predicted categories.

According to one embodiment, a magnetic reproduction processing device includes a decoder. The decoder includes a convolutional layer including a plurality of filters, and an attention layer configured to derive a degree of contribution related to the filters. The decoder is configured to output a decoded result obtained by integrating results of processing an input signal with the filters according to the degree of contribution.

A data storage device is disclosed comprising a head actuated over a disk comprising a data track having at least a first data segment and a second data segment. A first plurality of codewords are generated, and a first parity sector is generated over the first plurality of codewords. The first plurality of codewords and the first parity sector are written to the first data segment. A second plurality of codewords are generated, and a second parity sector is generated over the second plurality of codewords. The second plurality of codewords and the second parity sector are written to the second data segment. During a read operation the data segments of the data track are processed sequentially to decode the codewords using a low density parity check (LDPC) decoder, wherein the reliability metrics of un-converged codewords are stored in a codeword buffer and updated using the respective parity sector.

A data storage device is disclosed wherein a first plurality of codewords are generated each comprising a plurality of symbols, and a first parity sector is generated over the first plurality of codewords. A second plurality of codewords are generated each comprising a plurality of symbols, and a second parity sector is generated over the second plurality of codewords. A third parity sector is generated over a first subset of the first plurality of codewords and a first subset of the second plurality of codewords, and a fourth parity sector is generated over a second subset of the first plurality of codewords and a second subset of the second plurality of codewords. When processing of a first codeword fails, the first codeword and the first parity sector are processed using a LDPC type decoder, and the first codeword and the third parity sector are processed using the LDPC type decoder.

Technology is described herein for learning parameters for a parameterized iterative message passing decoder, and to a corresponding parameterized iterative message passing decoder. Learning the parameters may adapt the decoder to statistical dependencies introduced by the specific code's graph. Taking into account the statistical dependencies may allow the code to be shorter and/or denser. Note that the statistical dependencies in the graph may be extremely complex. Machine learning may be used to learn the parameters. The parameters may be learned when decoding data stored in the memory device. Learning the parameters may adapt the decoder to properties of data stored in the memory device, physical properties of the memory device, and/or patterns in host data.