Advanced Decoding Technics for Law Density Parity Check Codes Decoding

Abstract

Error-correcting codes based on very sparse matrices provide the best performance to date. We will focus on improving the decoding of Low Density Parity Checks Codes by examining a new variant of belief propagation for this problem. This new variant is based on techniques such as time averaging and prior biasing. Results obtained by extensive simulations show that, well combined, these techniques bring a modest improvement to BP decoding performance. We also studied the possibility of parallelism, which consists of breaking the algorithm in independent pieces that can be solved simultaneously on different computer nodes. But the time spent in inter-communication between the different machines slows the process and prevent the parallelisation to be as fast as expected.