EE Seminar: Diversified Decoding in HDPC Message Passing Algorithms

~~Nimrod Shacham, 
M.Sc. student under the supervision of Dr. Ofer Amrani

Wednesday, June 17, 2015 at 15:00
Room 011, Kitot Bldg., Faculty of Engineering

Diversified Decoding in HDPC Message Passing Algorithms

Abstract

For decades, Error Correcting Codes have been a fundamental building block of any communications system. The subject of ensuring good communication rate over unreliable channels is of critical importance and is becoming even more so for modern codes which push performance towards the theoretical limits of the communication channel. Low Density Parity Check Codes are one such family of codes. Despite the fact that they are lacking sound mathematical analysis, they have had tremendous success through the multitude of variations on the basic decoding algorithms and constructions. This thesis is concerned with employing the ideas behind iterative decoding of LDPC for other families of codes, particularly short codes with high density matrices (HDPC). There is a need for reliable graph-based iterative decoding of short codes; for instance this is beneficial in the framework of a newly introduced random version of OFDMA (R-OFDMA).
We first investigate how the structure of the parity check matrix can alter an iterative decoder's performance. Specifically we focus on the allowable redundancies in the matrices, where it is possible to benefit from inserting redundant parity checks without altering the code properties. We then investigate how altering the input LLRs to the decoder can affect the decoder's results and performance. By allowing multiple decoding passes, each with a slightly altered input matrix, one can positively affect error performance, keep the decoding latency unaltered and only increase the decoding complexity. Several algorithms for achieving this are offered and studied. We employ computation tree analysis and density evolution which support the simulation results.