Marvell International Limited

Facebook, Inc.

Marvellworld Trade Ltd.

Yifei Zhang

Gregory Burd

Nedeljko Varnica

Shu Li

Panu Chaichanavong

Hongwei Song

Mats Oberg

Kiran Kumar Gunnam

Nitin Nangare

Qiyue Zou

Wei Cao

Tianshi Gao

Dinkar Jain

Naim Siemsen-Schumann

Sina Jafarpour

Qian Yan

Satya Satyavarta

An online system maintains one or more models that determine likelihoods of a user performing various interactions after being presented with a content item. Additionally, the online system receives information identifying interactions by users with content, and generates embeddings for various users based on the interactions by the users with content. When determining whether to present a content item including an objective identifying an interaction to a user, the online system applies a maintained model to determine a likelihood of the user performing the interaction identified by the objective after being presented with the content item. Additionally, the online system determines a similarity of the embedding of the user to embeddings of users who performed the interaction identified by the objective. Based on a combination of the likelihood determined by the model and the similarity, the online system determines whether to present the content item to the user.

Accounting for organically occurring interactions with content when selecting content items for presenstation to users of an online system

Systems, methods, and apparatus are provided for iteratively decoding a codeword. Once a codeword is received, the codeword is processed to generate an incremental hard decision value and a log likelihood ratio amplitude value. These values are generated by processing the codeword using a soft output Viterbi algorithm. A faulty symbol in the codeword is identified. A complete hard decision value is generated using the incremental hard decision value. The LLR amplitude value and complete hard decision value corresponding to the identified faulty symbol are selectively provided to a decoder and the decoder uses these values to decode the codeword.

Implementation of LLR biasing method in non-binary iterative decoding

Various embodiments provide the position error signal (PES) that is generated by the data read channel that uses internal signals such as analog-to-digital signals, recovered data bits and signals or other internal signals to determine read head off-track metrics. Several different metrics can be utilized including a normalized norm metric, a normalized cross correlation metric, and a cross correlation with detected data metric. These metrics can be utilized to determine off-track shift of one or more read heads such that the read head position can be corrected.

Digital data driven position error signal for magnetic recording

Systems and methods for error recovery are presented. Data is decoded with an iterative decoding scheme having a first set of parameters. In response to a determination that the iterative decoding scheme has failed, the data is re-read. While the data is being re-read, the iterative decoding scheme is reconfigured with a second set of parameters, and the data is decoded with the reconfigured iterative decoding scheme. In response to determination that the reconfigured iterative decoding scheme has failed, an error type associated with the data is determined. An error recovery scheme is selected from a plurality of error recovery schemes for the data based on the determined error type.

Methods and apparatus for error recovery in memory systems employing iterative codes

In nonbinary iterative decoding, a data recovery scheme corrects for corrupted or defective data by determining reliability metrics for blocks of decoded nonbinary data. Block or windowed detectors generate block reliability metrics for data blocks (rather than individual bits) of decoded data using soft information from the regular decoding mode or from new iterative decoding iterations performed during defect detection mode. A defect detection system triggers corrective decoding of selected data blocks based on the block reliability metrics, by for example, comparing the block reliability metrics to a threshold or by selecting an adjustable number of the least reliable data blocks.

Methods and apparatus for defect detection and correction via nonbinary iterative decoding

Systems, methods, and apparatus are provided for improving the iterative decoding performance of a decoder, for example, as used in a wireless communications receiver or in a data retrieval unit. A decoding technique may receive and process a set of channel samples using an iterative decoder. If the iterative decoder output indicates a decoding failure, noise samples may foe combined with the received channel samples to create biased channel samples. Noise sample may be generated using a pseudo-random noise generator and/or by using signals already present in the communications receiver or data retrieval unit. The biased channel samples may be provided to the iterative decoder and the iterative decoder may re-run using the biased channel samples.

Iterative decoding systems using noise-biasing

Systems and methods are provided for enhancing the performance and throughput of a low-density parity check (LDPC) decoder. In some embodiments, the enhanced performance and throughput may be achieved by detecting and correcting near-codewords before the decoder iterates up to a predetermined number of iterations. In some embodiments, a corrector runs concurrently with the decoder to correct a near-codeword when the near-codeword is detected. In alternate embodiments, the corrector is active while the decoder is not active. Both embodiments allow for on-the-fly codeword error corrections that improve the performance (e.g., reducing the number of errors) without decreasing the throughput of the decoder.

Systems and methods for near-codeword detection and correction on the fly

A QC-LDPC decoding system employing a trapping set look-up table is provided. The QC-LDPC decoding system includes an iterative decoder that utilizes a message-passing algorithm to decode a received codeword. If the iterative decoder fails to produce a valid codeword, additional processing is performed to decode the received codeword. The additional processing includes the steps of computing the syndrome pattern of the received codeword, searching the look-up table for a trapping set class that is responsible for the iterative decoder's failure, retrieving from the look-up table a syndrome pattern and an error pattern of a member of the responsible trapping set class, and calculating the error pattern of the received codeword based on its syndrome pattern and the information retrieved from the look-up table. The received codeword is then corrected based on its error pattern.

QC-LDPC decoder with list-syndrome decoding

In iterative decoding, a data recovery scheme corrects for corrupted or defective data by determining reliability metrics for blocks of decoded data. Block or windowed detectors generate block reliability metrics for data blocks (rather than individual bits) of decoded data using soft information from the regular decoding mode or from new iterative decoding iterations performed during error recovery mode. An error recovery system triggers corrective decoding of selected data blocks based on the block reliability metrics, by for example, comparing the block reliability metrics to a threshold or by selecting an adjustable number of the least reliable data blocks.

Methods and apparatus for defect detection and correction via iterative decoding algorithms

The present disclosure includes systems and techniques relating to decoding signals produced within a storage device. A described technique includes retrieving a first codeword from a storage medium, decoding the first codeword, performing a retry process when the decoding was not successful, and retrieving one or more second codewords from the storage medium during the retry process to at least maintain a drive throughput. The retry process can include identifying one or more data chunks within the first codeword having potential defects, generating an erasure mask based on the one or more data chunks, applying, based on a window, one or more erasures within one or more different regions of the first codeword based on one or more corresponding regions of the erasure mask to produce one or more versions of the first codeword, and decoding the one or more versions of the first codeword.

High-throughput iterative decoding's defect scan in retry mode of storage system channel

Systems and methods for decoding a received codeword are provided. The received codeword is decoded based on a parity code to produce a plurality of checks. A first unsatisfied check is selected from the plurality of checks, and a first set of symbol positions in the received codeword that are connected to the first unsatisfied check is identified. A second set of symbol positions in the received codeword that are not connected to the first unsatisfied check is identified. The received codeword is modified by setting the first set of symbol positions to first predetermined values and by setting the second set of symbol positions to second predetermined values. The modified received codeword is decoded based on the parity code.

Post-processing methodologies in decoding LDPC codes

A QC-LDPC decoding system employing a trapping set look-up table is provided. The entries of the trapping set look-up table may be sorted according to failure frequencies of the trapping sets. The decoder may determine short-cycles associated with dominant trapping sets in order to decode the received codeword. If the iterative decoder of the QC-LDPC decoding system fails to produce a valid codeword, the decoder may compute the syndrome pattern of the processed codeword and search the look-up table for a trapping set class that is responsible for the iterative decoder's failure. If no responsible trapping set is found in the look-up table, the decoder may attempt to decode the received codeword using alternate decoding methods and subsequently determine a trapping set associated with the decoded codeword. If a trapping set is determined, then that trapping set may be added to the look-up table.

Growing community of inventors

Milpitas, CA, United States of America

Yifei Zhang