1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304
//! # MacKay-Neal pseudorandom LDPC construction
//!
//! This implements the algorithms from *MacKay, D.J. and Neal, R.M., 1996.
//! Near Shannon limit performance of low density parity check codes.
//! Electronics letters, 32(18), p.1645.* and variations on this idea.
//!
//! The algorithm works by adding column by column to the parity check
//! matrix. At each step, `wc` rows from the subset of rows that have not yet
//! achieved the total row weight `wr` are random chosen, and ones are inserted
//! in those positions.
//!
//! Optionally, to enforce a minimum girth, at each step the candidate
//! column is checked to see if it maintains the girth of the graph at or above
//! the minimum. If not, another random candidate column is
//! chosen according to the available rows. The algorithm aborts if after
//! a fixed number of trials it is unable to yield a new column satisfying
//! the required properties.
//!
//! # Examples
//! To run a MacKay-Neal LDPC generation algorithm, it is necessary
//! to create a [`Config`] and then use the `run()` method.
//! ```
//! # use ldpc_toolbox::mackay_neal::{Config, FillPolicy};
//! let conf = Config {
//! nrows: 4,
//! ncols: 8,
//! wr: 4,
//! wc: 2,
//! backtrack_cols: 0,
//! backtrack_trials: 0,
//! min_girth: None,
//! girth_trials: 0,
//! fill_policy: FillPolicy::Uniform,
//! };
//! let seed = 42;
//! let h = conf.run(seed).unwrap();
//! print!("{}", h.alist());
//! ```
use crate::rand::{Rng, *};
use crate::sparse::{Node, SparseMatrix};
use crate::util::*;
use rand::seq::IteratorRandom;
use rayon::prelude::*;
use std::fmt;
use std::fmt::{Display, Formatter};
/// Runtime errors of the MacKay-Neal construction.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum Error {
/// No rows available.
NoAvailRows,
/// Girth is too small (should not be returned to the user).
GirthTooSmall,
/// Exceeded backtrack trials.
NoMoreBacktrack,
/// Exceeded girth trials.
NoMoreTrials,
}
impl Display for Error {
fn fmt(&self, f: &mut Formatter<'_>) -> fmt::Result {
match self {
Error::NoAvailRows => write!(f, "no rows available"),
Error::GirthTooSmall => write!(f, "girth is too small"),
Error::NoMoreBacktrack => write!(f, "exceeded backtrack trials"),
Error::NoMoreTrials => write!(f, "exceeded girth trials"),
}
}
}
impl std::error::Error for Error {}
/// Result type used to indicate MacKay-Neal runtime errors.
pub type Result<T> = std::result::Result<T, Error>;
/// Configuration for the MacKay-Neal construction.
///
/// This configuration is used to set the parameters of the
/// LDPC code to construct as well as some options that affect
/// the exectution of the algorithm.
#[derive(Debug, Clone, PartialEq, Eq)]
pub struct Config {
/// Number of rows of the parity check matrix.
pub nrows: usize,
/// Number of columns of the parity check matrix.
pub ncols: usize,
/// Maximum row weight of the parity check matrix.
pub wr: usize,
/// Column weight of the parity check matrix.
pub wc: usize,
/// Number of columns to backtrack when there are not enough
/// available columns with weight smaller than the maximum row
/// weight.
pub backtrack_cols: usize,
/// Number of times to attempt backtracking before aborting.
pub backtrack_trials: usize,
/// Minimum girth of the Tanner graph; `None` indicates that
/// no constraints are imposed on the girth.
pub min_girth: Option<usize>,
/// Number of times to re-try generating a column to satisfy
/// the minimum girth constraint before aborting.
pub girth_trials: usize,
/// Policy used to select the rows to fill.
pub fill_policy: FillPolicy,
}
impl Config {
/// Runs the MacKay-Neal algorith using a random seed `seed`.
pub fn run(&self, seed: u64) -> Result<SparseMatrix> {
MacKayNeal::new(self, seed).run()
}
/// Searches for a seed for a successful MacKay-Neal construction
/// by trying several seeds.
///
/// The search is performed in parallel using a parallel iterator
/// from the rayon crate. This function returns the successful seed
/// and the corresponding parity check matrix, if one is found, or
/// `None` otherwise.
pub fn search(&self, start_seed: u64, max_tries: u64) -> Option<(u64, SparseMatrix)> {
(start_seed..start_seed + max_tries)
.into_par_iter()
.filter_map(|s| self.run(s).ok().map(|x| (s, x)))
.find_any(|_| true)
}
}
/// Policy used to select the rows to fill when adding a new column
/// in the MacKay-Neal algorith.
///
/// The `Random` policy chooses rows completely randomly, and only
/// imposes the maximum row weigth constraint of the configuration.
/// This has the drawback that near the end of the algorithm too
/// many rows can be full, while other rows have too many missing
/// items. Therefore, the algorithm will fail. Even if backtracking
/// is used, it is unlikely that the algorithm suceeds when the
/// matrix is large and the LDPC code is regular, so that all the
/// rows will necessarily end up with the same row weight.
///
/// The `Uniform` policy solves this problem by always picking
/// rows which have the lower weight possible. There is still some
/// randomness involved in the selection of rows, but the only
/// choices that are considered are those with the property that
/// it is not possible to exchange one of the choosen rows by
/// another row that was not chosen and has stricly less weight.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum FillPolicy {
/// Choose randomly from the set of rows whose weight is less
/// than the maximum row weight.
Random,
/// Try to choose only among the rows which have lower weight.
Uniform,
}
struct MacKayNeal {
wr: usize,
wc: usize,
h: SparseMatrix,
rng: Rng,
backtrack_cols: usize,
backtrack_trials: usize,
min_girth: Option<usize>,
girth_trials: usize,
fill_policy: FillPolicy,
current_col: usize,
}
impl MacKayNeal {
fn new(conf: &Config, seed: u64) -> MacKayNeal {
MacKayNeal {
wr: conf.wr,
wc: conf.wc,
h: SparseMatrix::new(conf.nrows, conf.ncols),
rng: Rng::seed_from_u64(seed),
backtrack_cols: conf.backtrack_cols,
backtrack_trials: conf.backtrack_trials,
min_girth: conf.min_girth,
girth_trials: conf.girth_trials,
fill_policy: conf.fill_policy,
current_col: 0,
}
}
fn try_insert_column(&mut self) -> Result<()> {
let rows = self.select_rows()?;
self.h.insert_col(self.current_col, rows.into_iter());
if let Some(g) = self.min_girth {
if self
.h
.girth_at_node_with_max(Node::Col(self.current_col), g - 1)
.is_some()
{
self.h.clear_col(self.current_col);
return Err(Error::GirthTooSmall);
}
}
Ok(())
}
fn select_rows(&mut self) -> Result<Vec<usize>> {
match self.fill_policy {
FillPolicy::Random => {
let h = &self.h;
let wr = self.wr;
let avail_rows = (0..self.h.num_rows()).filter(|&r| h.row_weight(r) < wr);
let select_rows = avail_rows.choose_multiple(&mut self.rng, self.wc);
if select_rows.len() < self.wc {
return Err(Error::NoAvailRows);
}
Ok(select_rows)
}
FillPolicy::Uniform => {
let avail_rows: Vec<(usize, usize)> = (0..self.h.num_rows())
.filter_map(|r| {
let w = self.h.row_weight(r);
if w < self.wr {
Some((r, w))
} else {
None
}
})
.collect();
avail_rows
.sort_by_random_sel(self.wc, |(_, x), (_, y)| x.cmp(y), &mut self.rng)
.map(|a| a.into_iter().map(|(x, _)| x).collect())
.ok_or(Error::NoAvailRows)
}
}
}
fn backtrack(&mut self) -> Result<()> {
if self.backtrack_trials == 0 {
return Err(Error::NoMoreBacktrack);
}
self.backtrack_trials -= 1;
let b = std::cmp::min(self.current_col, self.backtrack_cols);
let a = self.current_col - b;
for col in a..self.current_col {
self.h.clear_col(col);
}
self.current_col = a;
Ok(())
}
fn retry_girth(&mut self) -> Result<()> {
if self.girth_trials == 0 {
return Err(Error::NoMoreTrials);
}
self.girth_trials -= 1;
Ok(())
}
fn run(mut self) -> Result<SparseMatrix> {
while self.current_col < self.h.num_cols() {
match self.try_insert_column() {
Ok(_) => self.current_col += 1,
Err(Error::NoAvailRows) => self.backtrack()?,
Err(Error::GirthTooSmall) => self.retry_girth()?,
Err(e) => return Err(e),
};
}
Ok(self.h)
}
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn small_matrix() {
let conf = Config {
nrows: 4,
ncols: 8,
wr: 4,
wc: 2,
backtrack_cols: 0,
backtrack_trials: 0,
min_girth: None,
girth_trials: 0,
fill_policy: FillPolicy::Random,
};
let h = conf.run(187).unwrap();
let alist = "8 4
2 4
2 2 2 2 2 2 2 2
4 4 4 4
1 3
3 4
1 4
1 4
1 2
2 3
2 3
2 4
1 3 4 5
5 6 7 8
1 2 6 7
2 3 4 8
";
assert_eq!(h.alist(), alist);
}
}