sciroccogti/ldpc-with-simd

LDPC with SIMD aimed at AMD CPUs.

LDPC-with-SIMD

LDPC with SIMD.

⚠️ Just a experiment project, currently it's not using SIMD, and may not work properly.

Environment

• OS: Ubuntu 20.04.1 or 18.04.4
• Instruction: AVX2 or AVX512
• Dependency:
  • CMake
  • Conon
  • Eigen3
  • Python3-dev

Functions

Binary LDPC

• Encode
• Modulate(BPSK)
• Channel(AWGN)
• Demodulate(BPSK)
• Decode(NMS/SPA)

NonBinary LDPC

• Encode
• Modulate(BPSK)
• Channel(AWGN)
• Demodulate(BPSK)
• Decode(EMS)

Tools

• Draw Tanner
• Type in Alist

Install Conan

sudo apt install python3-pip
pip3 install conan

Install Python3-dev

sudo apt install python3-dev

How to use

1 Simulation

mkdir build
cd build
conan install .. --build=yaml-cpp
conan profile update settings.compiler.libcxx=libstdc++11 default
cmake -DBUILD_SHARED_LIBS=ON ..
cmake --build .
../bin/LDPC-with-SIMD --dec-h-path ../example/H.alist

2 Tools

2-1 Draw Tanner

python3 script/tanner.py --dec-h-path example/H.alist

2-1 Type in Alist

python3 script/makeAlist.py
# nRow is: 4
# nCol is: 6
# 1 1 0 1 0 0 
# 0 1 1 0 1 0
# 1 0 0 0 1 1
# 0 0 1 1 0 1
# filename is: aaa.alist

References

Binary LDPC and Basic

• pyldpc
• 5G-SIMD-LDPC
• Python处理alist文件
• aff3ct
• paper: NMS

Nonbinary LDPC

• NB_LDPC_FB
• Kaiserslautern database

C++ tutorial

• c++模板类(一)理解编译器的编译模板过程
• Conan-CMake transparent integration

ldpc-with-simd's Issues

Speed comparison between LUT and on-the-fly calculated GF implement

LUT implement is using arrays or something similar to store calculation result in advance.

on-the-fly calculated implement is something like libff's implement

really should accumulate in matrix use xor?!

https://github.com/Sciroccogti/LDPC-with-SIMD/blob/495722103f22dd1da65aa640add34aa75aa7b077/src/MatrixMath/NBMatrixMath.cpp#L185-L185

issues on Encoder

• support H which has a full-zero row
• support H which is singular

Notes

How to detect the AVX supported by your platform?

gcc -march=native -dM -E - < /dev/null | egrep "SSE|AVX" | sort

False SPA implement in Binary LDPC

You stupid missed tentative output!

support nonSystematic NBLDPC

LDPC-with-SIMD/src/LDPC/NBLDPC.cpp

Line 147 in f7e5750

}

Real BPSK will introduce error

LDPC-with-SIMD/src/Modem/Modem.cpp

Line 51 in e526358