acg-alp-ldpc

Implementation of LDPC codes decoding algorithms:

• Belief Propagation
• Adaptive Cut Generation Algorithm
• QP-ADMM Algorithm

Also implementation of the optimization of check matrix for QP-ADMM algorithm.

Installation

To run Linear Programming based methods, glpk library needs to be installed. Installation instructions (works only on mac):

git clone https://github.com/GreatDrake/acg-alp-ldpc.git && cd acg-alp-ldpc
wget http://ftp.gnu.org/gnu/glpk/glpk-5.0.tar.gz
tar -xf glpk-5.0.tar.gz && cd glpk-5.0
./configure && make
mv src/glpk.h ../ && mv src/.libs/libglpk.a ../ && cd ../

On Ubuntu one can evaluate Belief Propagation and QP-ADMM algorithms without installing glpk. For that, after installation above run:

sed -i '1 s/^/\/\//' main.cpp

Run experiments

Algorithms

To evaluate algorithms on matrix at data/optimalH.txt, run:

make run

This command will output algorithm results to stdout and report.csv.

Matrix optimization

To run local decsent algorithm for check matrix optimization for QP-ADMM algorithm (see its parameters in optimize_H.cpp), run:

make optimize

This command will overwrite file data/optimalH.txt. Adjust constant THREADS_NUM in optimize_H.cpp before run.

QP-ADMM parameters optimization

To run grid-search on QP-ADMM parameters for matrix at data/optimalH.txt, run:

make run_qpadmm_params

This command will output optimal alpha and mu to stdout.