diff --git a/srctmp/channel.rs b/srctmp/channel.rs
new file mode 100644
index 0000000..c033cf9
--- /dev/null
+++ b/srctmp/channel.rs
@@ -0,0 +1,192 @@
+use crate::{Gf2, LdpcError, Llr, Result};
+
+// Trait Channel
+
+pub trait Channel: Send + Sync {
+    fn transmit(&self, codeword: &[Gf2], rng: &mut impl rand::Rng) -> Vec<Llr>;
+    fn capacity(&self) -> f64;
+}
+
+// Canal AWGN
+// Modulation BPSK : 0 -> +1.0, 1 -> -1.0
+// Signal reçu : y = x + n, n eq N(0, sig²)
+// LLR optimal : L(y) = 2y/sig²
+// sig² = N_0/2 = 1/(2 R SNR_lin)
+
+#[derive(Debug, Clone)]
+pub struct AwgnChannel {
+    pub snr_db: f64,
+    sigma: f64,
+}
+
+impl AwgnChannel {
+    pub fn new(snr_db: f64, code_rate: f64) -> Result<Self> {
+        if !(0.0..1.0).contains(&code_rate) {
+            return Err(LdpcError::OutOfRange("code_rate ∈ ]0, 1[".into()));
+        }
+        let snr_lin = 10.0_f64.powf(snr_db / 10.0);
+        let sigma = (1.0 / (2.0 * code_rate * snr_lin)).sqrt();
+        Ok(Self { snr_db, sigma })
+    }
+
+    pub fn sigma(&self) -> f64 {
+        self.sigma
+    }
+    pub fn snr_linear(&self) -> f64 {
+        10.0_f64.powf(self.snr_db / 10.0)
+    }
+
+    #[inline]
+    pub fn llr_from_received(y: f64, sigma: f64) -> Llr {
+        2.0 * y / (sigma * sigma)
+    }
+}
+
+impl Channel for AwgnChannel {
+    fn transmit(&self, codeword: &[Gf2], rng: &mut impl rand::Rng) -> Vec<Llr> {
+        use rand_distr::{Distribution, Normal};
+        let normal = Normal::new(0.0, self.sigma).unwrap();
+        codeword
+            .iter()
+            .map(|&b| {
+                let x = if b == 0 { 1.0_f64 } else { -1.0_f64 };
+                let y = x + normal.sample(rng);
+                Self::llr_from_received(y, self.sigma)
+            })
+            .collect()
+    }
+
+    fn capacity(&self) -> f64 {
+        // Capacité BPSK-AWGN par Monte-Carlo
+        use rand_distr::{Distribution, Normal};
+        let mut rng = rand::thread_rng();
+        let normal = Normal::new(0.0, self.sigma).unwrap();
+        let n_samples = 10_000usize;
+        let mut sum = 0.0f64;
+        for _ in 0..n_samples {
+            let n: f64 = normal.sample(&mut rng);
+            let y = 1.0 + n; // bit 0 transmis (x=+1)
+            let llr = Self::llr_from_received(y, self.sigma);
+            // I = 1 - E[log2(1 + exp(-2y/sig²))]
+            sum += (1.0 + (-llr).exp()).log2();
+        }
+        1.0 - sum / n_samples as f64
+    }
+}
+
+// Canal BSC
+// Chaque bit inversé avec probabilité p
+// LLR : +-log((1-p)/p) selon le bit reçu
+
+#[derive(Debug, Clone)]
+pub struct BscChannel {
+    pub crossover_prob: f64,
+    llr_magnitude: Llr,
+}
+
+impl BscChannel {
+    pub fn new(crossover_prob: f64) -> Result<Self> {
+        if crossover_prob <= 0.0 || crossover_prob >= 0.5 {
+            return Err(LdpcError::OutOfRange("p ∈ ]0, 0.5[".into()));
+        }
+        let llr_magnitude = ((1.0 - crossover_prob) / crossover_prob).ln();
+        Ok(Self {
+            crossover_prob,
+            llr_magnitude,
+        })
+    }
+}
+
+impl Channel for BscChannel {
+    fn transmit(&self, codeword: &[Gf2], rng: &mut impl rand::Rng) -> Vec<Llr> {
+        codeword
+            .iter()
+            .map(|&b| {
+                let rcv = if rng.gen::<f64>() < self.crossover_prob {
+                    b ^ 1
+                } else {
+                    b
+                };
+                if rcv == 0 {
+                    self.llr_magnitude
+                } else {
+                    -self.llr_magnitude
+                }
+            })
+            .collect()
+    }
+
+    // C_BSC(p) = 1 - Hb(p) avec Hb = entropie binaire
+    fn capacity(&self) -> f64 {
+        let p = self.crossover_prob;
+        let hb = -p * p.log2() - (1.0 - p) * (1.0 - p).log2();
+        1.0 - hb
+    }
+}
+
+// Canal BEC
+// Bit effacé avec probabilité ε -> LLR = 0 (incertitude totale)
+// Bit reçu correctement → LLR = +-CERTAIN_LLR (grand mais fini)
+
+#[derive(Debug, Clone)]
+pub struct BecChannel {
+    pub erasure_prob: f64,
+}
+
+impl BecChannel {
+    pub fn new(erasure_prob: f64) -> Result<Self> {
+        if erasure_prob <= 0.0 || erasure_prob >= 1.0 {
+            return Err(LdpcError::OutOfRange("ε ∈ ]0, 1[".into()));
+        }
+        Ok(Self { erasure_prob })
+    }
+
+    // Grand mais fini pour stabilité numérique
+    const CERTAIN_LLR: Llr = 100.0;
+}
+
+impl Channel for BecChannel {
+    fn transmit(&self, codeword: &[Gf2], rng: &mut impl rand::Rng) -> Vec<Llr> {
+        codeword
+            .iter()
+            .map(|&b| {
+                if rng.gen::<f64>() < self.erasure_prob {
+                    0.0 // Effacement
+                } else if b == 0 {
+                    Self::CERTAIN_LLR
+                } else {
+                    -Self::CERTAIN_LLR
+                }
+            })
+            .collect()
+    }
+
+    /// C_BEC(ε) = 1 - ε
+    fn capacity(&self) -> f64 {
+        1.0 - self.erasure_prob
+    }
+}
+
+// #[cfg(test)]
+// mod tests {
+//     use super::*;
+//
+//     #[test]
+//     fn test_awgn_llr_formula() {
+//         assert!((AwgnChannel::llr_from_received(1.0, 1.0) - 2.0).abs() < 1e-12);
+//         assert!((AwgnChannel::llr_from_received(-1.0, 1.0) + 2.0).abs() < 1e-12);
+//     }
+//
+//     #[test]
+//     fn test_bec_capacity() {
+//         let bec = BecChannel::new(0.3).unwrap();
+//         assert!((bec.capacity() - 0.7).abs() < 1e-12);
+//     }
+//
+//     #[test]
+//     fn test_bsc_capacity_bounds() {
+//         let bsc = BscChannel::new(0.1).unwrap();
+//         let cap = bsc.capacity();
+//         assert!(cap > 0.0 && cap < 1.0);
+//     }
+// }
diff --git a/srctmp/code.rs b/srctmp/code.rs
new file mode 100644
index 0000000..3478088
--- /dev/null
+++ b/srctmp/code.rs
@@ -0,0 +1,453 @@
+use crate::graph::TannerGraph;
+use crate::matrix::{DenseMatrixGF2, SparseMatrixGF2};
+use crate::{Gf2, LdpcError, Result};
+
+#[derive(Debug, Clone)]
+pub struct LdpcParams {
+    pub n: usize,
+    pub k: usize,
+    pub topology: CodeTopology,
+    pub generation: GenerationMethod,
+    pub seed: Option<u64>,
+}
+
+impl LdpcParams {
+    pub fn rate(&self) -> f64 {
+        self.k as f64 / self.n as f64
+    }
+    pub fn m(&self) -> usize {
+        self.n - self.k
+    }
+
+    pub fn validate(&self) -> Result<()> {
+        if self.k >= self.n {
+            return Err(LdpcError::InvalidParameters("k doit être < n".into()));
+        }
+        if self.n < 4 {
+            return Err(LdpcError::InvalidParameters("n trop petit".into()));
+        }
+        self.topology.validate(self)
+    }
+}
+
+// Topologie
+
+#[derive(Debug, Clone)]
+pub enum CodeTopology {
+    // Chaque var-node a degré wc, chaque check-node a degré wr
+    // Condition nécessaire : n * wc == m * wr.
+    Regular { wc: usize, wr: usize },
+    // Degrés définis par des polynômes de distribution (density evolution)
+    // lambda[i] = fraction d'arêtes sur des var-nodes de degré i+2
+    // rho[i]   = fraction d'arêtes sur des check-nodes de degré i+2
+    Irregular { lambda: Vec<f64>, rho: Vec<f64> },
+}
+
+impl CodeTopology {
+    fn validate(&self, params: &LdpcParams) -> Result<()> {
+        match self {
+            CodeTopology::Regular { wc, wr } => {
+                if params.n * wc != params.m() * wr {
+                    return Err(LdpcError::InvalidParameters(format!(
+                        "n*wc ({}) doit égaler m*wr ({}) pour un code régulier",
+                        params.n * wc,
+                        params.m() * wr
+                    )));
+                }
+            }
+            CodeTopology::Irregular { lambda, rho } => {
+                let sl: f64 = lambda.iter().sum();
+                let sr: f64 = rho.iter().sum();
+                if (sl - 1.0).abs() > 1e-6 || (sr - 1.0).abs() > 1e-6 {
+                    return Err(LdpcError::InvalidParameters(
+                        "Les polynômes lambda et rho doivent sommer à 1".into(),
+                    ));
+                }
+            }
+        }
+        Ok(())
+    }
+}
+
+// Méthode de génération
+
+#[derive(Debug, Clone)]
+pub enum GenerationMethod {
+    // H = [H1 | H2 | ... | Hwc]^T, H1 régulière, Hi = permutation de H1
+    Gallager,
+    // Ajout de colonnes de poids fixe, rejet si cycle-4 créé
+    MacKayNeal { max_attempts: usize },
+    // Progressive Edge Growth : maximise le girth local arête par arête
+    Peg,
+}
+
+// Forme systématique
+
+#[derive(Debug, Clone)]
+pub struct SystematicForm {
+    // G = [I_k | P], dense car P est généralement pleine
+    pub g: DenseMatrixGF2,
+    // Permutation de colonnes appliquée à H pour obtenir [A | I_m]
+    pub col_perm: Vec<usize>,
+    // Permutation inverse pour réordonner le mot de code
+    pub col_perm_inv: Vec<usize>,
+}
+
+// Structure principale
+
+#[derive(Debug, Clone)]
+pub struct LdpcCode {
+    pub params: LdpcParams,
+    pub h: SparseMatrixGF2,
+    pub graph: TannerGraph,
+    pub systematic_form: Option<SystematicForm>,
+}
+
+impl LdpcCode {
+    pub fn new(params: LdpcParams) -> Result<Self> {
+        params.validate()?;
+        let mut rng = build_rng(params.seed);
+        let h = match &params.generation {
+            GenerationMethod::Gallager => generate_gallager(&params, &mut rng)?,
+            GenerationMethod::MacKayNeal { max_attempts } => {
+                generate_mackay_neal(&params, *max_attempts, &mut rng)?
+            }
+            GenerationMethod::Peg => generate_peg(&params)?,
+        };
+        let graph = TannerGraph::from_matrix(&h);
+        Ok(Self {
+            params,
+            h,
+            graph,
+            systematic_form: None,
+        })
+    }
+
+    pub fn from_matrix(h: SparseMatrixGF2, k: usize) -> Result<Self> {
+        let n = h.cols;
+        let params = LdpcParams {
+            n,
+            k,
+            topology: CodeTopology::Regular { wc: 0, wr: 0 }, // inconnu
+            generation: GenerationMethod::Gallager,
+            seed: None,
+        };
+        let graph = TannerGraph::from_matrix(&h);
+        Ok(Self {
+            params,
+            h,
+            graph,
+            systematic_form: None,
+        })
+    }
+
+    // Calcule G par Gauss-Jordan sur H (to cache)
+    // Appelé automatiquement par SystematicEncoder si nécessaire
+    pub fn compute_systematic_form(&mut self) -> Result<()> {
+        if self.systematic_form.is_some() {
+            return Ok(());
+        }
+
+        let dense = self.h.to_dense();
+        // On travaille sur [H | I_m] pour tracker les opérations de lignes
+        let m = self.params.m();
+        let n = self.params.n;
+
+        // Construire la matrice augmentée de taille m (n + m)
+        let mut aug = DenseMatrixGF2::zeros(m, n + m);
+        for r in 0..m {
+            for c in 0..n {
+                aug.set(r, c, dense[r][c]);
+            }
+            aug.set(r, n + r, 1); // partie identité
+        }
+
+        // Gauss-Jordan sur les n premières colonnes
+        let (col_perm, rank) = aug.gauss_jordan_gf2();
+        if rank < m {
+            return Err(LdpcError::SingularMatrix);
+        }
+
+        // Extraire P (colonnes n..n+m de aug) -> partie redondante de G
+        // G = [I_k | P^T] après transposition correcte
+        // TODO: extraire proprement P et construire G = [I_k | P]
+        let g = DenseMatrixGF2::identity(self.params.k); // placeholder
+
+        let col_perm_inv = {
+            let mut inv = vec![0usize; col_perm.len()];
+            for (i, &p) in col_perm.iter().enumerate() {
+                inv[p] = i;
+            }
+            inv
+        };
+
+        self.systematic_form = Some(SystematicForm {
+            g,
+            col_perm,
+            col_perm_inv,
+        });
+        Ok(())
+    }
+
+    pub fn rate(&self) -> f64 {
+        self.params.rate()
+    }
+    pub fn n(&self) -> usize {
+        self.params.n
+    }
+    pub fn k(&self) -> usize {
+        self.params.k
+    }
+    pub fn m(&self) -> usize {
+        self.params.m()
+    }
+    pub fn girth(&self) -> usize {
+        self.graph.girth()
+    }
+
+    pub fn is_codeword(&self, c: &[Gf2]) -> bool {
+        self.h.multiply_vec(c).iter().all(|&s| s == 0)
+    }
+}
+
+// RNG
+
+fn build_rng(seed: Option<u64>) -> impl rand::Rng {
+    use rand::SeedableRng;
+    rand::rngs::StdRng::seed_from_u64(seed.unwrap_or_else(rand::random))
+}
+
+// Gallager
+// H divisée en wc sous-matrices de taille (m/wc) n
+// H1 est une matrice régulière (chaque ligne contient exactement wr uns)
+// H2..Hwc sont des permutations aléatoires de colonnes de H1
+
+fn generate_gallager(params: &LdpcParams, rng: &mut impl rand::Rng) -> Result<SparseMatrixGF2> {
+    let CodeTopology::Regular { wc, wr } = params.topology else {
+        return Err(LdpcError::InvalidParameters(
+            "Gallager nécessite un code régulier".into(),
+        ));
+    };
+    let n = params.n;
+    let m = params.m();
+    if m % wc != 0 {
+        return Err(LdpcError::InvalidParameters(
+            "m doit être divisible par wc".into(),
+        ));
+    }
+    let rows_per_block = m / wc;
+    let mut ones: Vec<(usize, usize)> = Vec::new();
+
+    // H1 : blocs réguliers de wr uns par ligne
+    for r in 0..rows_per_block {
+        for j in 0..wr {
+            ones.push((r, r * wr + j));
+        }
+    }
+    // H2..Hwc : permutations aléatoires de colonnes
+    use rand::seq::SliceRandom;
+    for block in 1..wc {
+        let mut perm: Vec<usize> = (0..n).collect();
+        perm.shuffle(rng);
+        for r in 0..rows_per_block {
+            for j in 0..wr {
+                ones.push((block * rows_per_block + r, perm[r * wr + j]));
+            }
+        }
+    }
+    Ok(SparseMatrixGF2::from_positions(m, n, ones))
+}
+
+// MacKay-Neal
+// Ajoute les colonnes une à une avec poids wc
+// Rejette toute colonne créant un cycle-4 (deux colonnes ne peuvent
+// partager qu'un seul 1 commun). Relance si max_attempts est dépassé
+
+fn generate_mackay_neal(
+    params: &LdpcParams,
+    max_attempts: usize,
+    rng: &mut impl rand::Rng,
+) -> Result<SparseMatrixGF2> {
+    let CodeTopology::Regular { wc, .. } = params.topology else {
+        return Err(LdpcError::InvalidParameters(
+            "MacKayNeal nécessite un code régulier".into(),
+        ));
+    };
+    let n = params.n;
+    let m = params.m();
+    let mut ones: Vec<(usize, usize)> = Vec::new();
+
+    use rand::seq::SliceRandom;
+    for col in 0..n {
+        let mut placed = false;
+        for _attempt in 0..max_attempts {
+            // Tirer wc lignes sans remise
+            let mut rows: Vec<usize> = (0..m).collect();
+            rows.shuffle(rng);
+            let candidate: Vec<usize> = rows[..wc].to_vec();
+
+            // Vérifier cycle-4 : cette colonne ne partage pas ≥2 lignes avec une colonne existante
+            let mut ok = true;
+            let mut c2 = 0;
+            while c2 < col && ok {
+                let existing: Vec<usize> = ones
+                    .iter()
+                    .filter(|&&(_, c)| c == c2)
+                    .map(|&(r, _)| r)
+                    .collect();
+                let shared = candidate.iter().filter(|r| existing.contains(r)).count();
+                if shared >= 2 {
+                    ok = false;
+                }
+                c2 += 1;
+            }
+            if ok {
+                for r in candidate {
+                    ones.push((r, col));
+                }
+                placed = true;
+                break;
+            }
+        }
+        if !placed {
+            return Err(LdpcError::GenerationFailed {
+                attempts: max_attempts,
+            });
+        }
+    }
+    Ok(SparseMatrixGF2::from_positions(m, n, ones))
+}
+
+// PEG
+// Progressive Edge Growth
+// Pour chaque arête à ajouter, choisit le check-node qui maximise
+// le girth local du graphe courant (BFS depuis le var-node courant)
+
+fn generate_peg(params: &LdpcParams) -> Result<SparseMatrixGF2> {
+    let CodeTopology::Regular { wc, .. } = params.topology else {
+        return Err(LdpcError::InvalidParameters(
+            "PEG nécessite un code régulier".into(),
+        ));
+    };
+    let n = params.n;
+    let m = params.m();
+    let mut var_to_chk: Vec<Vec<usize>> = vec![vec![]; n];
+    let mut chk_to_var: Vec<Vec<usize>> = vec![vec![]; m];
+    let mut ones: Vec<(usize, usize)> = Vec::new();
+
+    for v in 0..n {
+        for _edge in 0..wc {
+            // BFS depuis v dans le graphe courant pour trouver le check-node
+            // non-voisin de v le plus éloigné (maximise le girth local)
+            let best_chk = peg_find_best_check(v, &var_to_chk, &chk_to_var, m);
+            var_to_chk[v].push(best_chk);
+            chk_to_var[best_chk].push(v);
+            ones.push((best_chk, v));
+        }
+    }
+    Ok(SparseMatrixGF2::from_positions(m, n, ones))
+}
+
+// BFS depuis le var-node v, retourne le check-node non-voisin de v
+// qui est le plus éloigné (ou le moins chargé en cas d'égalité)
+fn peg_find_best_check(
+    v: usize,
+    var_to_chk: &[Vec<usize>],
+    chk_to_var: &[Vec<usize>],
+    n_chk: usize,
+) -> usize {
+    use std::collections::VecDeque;
+    let current_neighbors: &[usize] = &var_to_chk[v];
+    let mut dist_chk = vec![usize::MAX; n_chk];
+    let mut dist_var = vec![usize::MAX; var_to_chk.len()];
+    dist_var[v] = 0;
+    let mut queue: VecDeque<(bool, usize, usize)> = VecDeque::new();
+    queue.push_back((true, v, 0));
+    let mut max_dist = 0;
+    let mut reachable_non_neighbors: Vec<(usize, usize)> = Vec::new(); // (dist, chk)
+
+    while let Some((is_var, node, dist)) = queue.pop_front() {
+        if is_var {
+            for &c in &var_to_chk[node] {
+                if dist_chk[c] == usize::MAX {
+                    dist_chk[c] = dist + 1;
+                    max_dist = max_dist.max(dist + 1);
+                    queue.push_back((false, c, dist + 1));
+                }
+            }
+        } else {
+            for &u in &chk_to_var[node] {
+                if dist_var[u] == usize::MAX {
+                    dist_var[u] = dist + 1;
+                    queue.push_back((true, u, dist + 1));
+                }
+            }
+        }
+    }
+
+    // Parmi les check-nodes non-voisins de v, choisir le plus éloigné
+    // (ou le moins chargé en cas d'égalité)
+    let mut best = (0usize, 0usize, usize::MAX); // (dist, chk_id, charge)
+    for c in 0..n_chk {
+        if current_neighbors.contains(&c) {
+            continue;
+        }
+        let d = dist_chk[c];
+        let load = chk_to_var[c].len();
+        if d > best.0 || (d == best.0 && load < best.2) {
+            best = (d, c, load);
+        }
+    }
+    // Fallback : check le moins chargé si aucun atteignable
+    if best.0 == 0 {
+        (0..n_chk).min_by_key(|&c| chk_to_var[c].len()).unwrap_or(0)
+    } else {
+        best.1
+    }
+}
+
+// #[cfg(test)]
+// mod tests {
+//     use super::*;
+//
+//     #[test]
+//     fn test_gallager_regular_degrees() {
+//         let params = LdpcParams {
+//             n: 12,
+//             k: 6,
+//             topology: CodeTopology::Regular { wc: 2, wr: 4 },
+//             generation: GenerationMethod::Gallager,
+//             seed: Some(42),
+//         };
+//         let code = LdpcCode::new(params).unwrap();
+//         for col in 0..code.n() {
+//             assert_eq!(code.h.col_weight(col), 2);
+//         }
+//     }
+//
+//     #[test]
+//     fn test_mackay_neal_no_4cycle() {
+//         let params = LdpcParams {
+//             n: 20,
+//             k: 10,
+//             topology: CodeTopology::Regular { wc: 3, wr: 6 },
+//             generation: GenerationMethod::MacKayNeal { max_attempts: 1000 },
+//             seed: Some(0),
+//         };
+//         let code = LdpcCode::new(params).unwrap();
+//         assert!(!code.graph.has_4_cycle());
+//     }
+//
+//     #[test]
+//     fn test_peg_girth_at_least_6() {
+//         let params = LdpcParams {
+//             n: 30,
+//             k: 15,
+//             topology: CodeTopology::Regular { wc: 3, wr: 6 },
+//             generation: GenerationMethod::Peg,
+//             seed: None,
+//         };
+//         let code = LdpcCode::new(params).unwrap();
+//         assert!(code.girth() >= 6);
+//     }
+// }
diff --git a/srctmp/decoder.rs b/srctmp/decoder.rs
new file mode 100644
index 0000000..14370e1
--- /dev/null
+++ b/srctmp/decoder.rs
@@ -0,0 +1,446 @@
+use crate::code::LdpcCode;
+use crate::graph::TannerGraph;
+use crate::matrix::SparseMatrixGF2;
+use crate::{BitVec, Gf2, Llr};
+
+// Résultat
+
+#[derive(Debug, Clone)]
+pub enum DecoderResult {
+    Converged(BitVec),
+    MaxIterationsReached(BitVec),
+    Failure,
+}
+
+impl DecoderResult {
+    pub fn codeword(&self) -> Option<&BitVec> {
+        match self {
+            DecoderResult::Converged(c) | DecoderResult::MaxIterationsReached(c) => Some(c),
+            DecoderResult::Failure => None,
+        }
+    }
+    pub fn is_success(&self) -> bool {
+        matches!(self, DecoderResult::Converged(_))
+    }
+}
+
+// Configuration
+
+#[derive(Debug, Clone)]
+pub struct DecoderConfig {
+    pub max_iterations: usize,
+    pub early_stopping: bool,
+}
+
+impl Default for DecoderConfig {
+    fn default() -> Self {
+        Self {
+            max_iterations: 50,
+            early_stopping: true,
+        }
+    }
+}
+
+// Trait Decoder
+
+pub trait Decoder: Send + Sync {
+    fn decode(&self, channel_llr: &[Llr]) -> DecoderResult;
+
+    fn decode_hard(&self, received: &[Gf2]) -> DecoderResult {
+        let llr: Vec<Llr> = received
+            .iter()
+            .map(|&b| if b == 0 { 1.0 } else { -1.0 })
+            .collect();
+        self.decode(&llr)
+    }
+}
+
+// Primitives GF(2) et LLR
+
+#[inline]
+pub fn hard_decision(llr: Llr) -> Gf2 {
+    if llr >= 0.0 {
+        0
+    } else {
+        1
+    }
+}
+
+pub fn compute_syndrome(h: &SparseMatrixGF2, c: &[Gf2]) -> Vec<Gf2> {
+    h.multiply_vec(c)
+}
+
+// phi(x) = -ln(tanh(|x|/2)) involution de Sum-Product
+// phi(phi(x)) = x
+#[inline]
+fn phi(x: Llr) -> Llr {
+    let ax = x.abs().max(1e-10);
+    -((ax / 2.0).tanh().ln())
+}
+
+// Mises à jour des nœuds
+
+// Mise à jour Sum-Product du nœud de contrôle
+// R_{c→v} = φ(∑_{v'≠v} φ(|L_{v'→c}|)) × sign(∏_{v'≠v} sign(L_{v'→c}))
+fn check_node_update_sp(incoming: &[Llr], out: &mut [Llr]) {
+    let phi_sum: Llr = incoming.iter().map(|&l| phi(l.abs())).sum();
+    let sign_prod: Llr = incoming.iter().map(|&l| l.signum()).product();
+    for (j, (&l, r)) in incoming.iter().zip(out.iter_mut()).enumerate() {
+        let phi_excl = phi_sum - phi(l.abs());
+        let sign_excl = sign_prod * l.signum();
+        *r = sign_excl * phi(phi_excl);
+    }
+}
+
+// Mise à jour Min-Sum avec facteur de normalisation α
+// Approxime φ(∑ φ(|L|)) ≈ min(|L|).
+// alpha in [0.75, 0.875] compense le biais de Min-Sum brut
+fn check_node_update_ms(incoming: &[Llr], out: &mut [Llr], alpha: Llr) {
+    let sign_prod: Llr = incoming.iter().map(|&l| l.signum()).product();
+    // Précalcul des deux plus petites valeurs absolues
+    let mut min1 = Llr::INFINITY;
+    let mut min2 = Llr::INFINITY;
+    let mut min1_idx = 0;
+    for (j, &l) in incoming.iter().enumerate() {
+        let al = l.abs();
+        if al < min1 {
+            min2 = min1;
+            min1 = al;
+            min1_idx = j;
+        } else if al < min2 {
+            min2 = al;
+        }
+    }
+    for (j, (&l, r)) in incoming.iter().zip(out.iter_mut()).enumerate() {
+        let min_excl = if j == min1_idx { min2 } else { min1 };
+        let sign_excl = sign_prod * l.signum();
+        *r = alpha * sign_excl * min_excl;
+    }
+}
+
+// Mise à jour du nœud de variable
+// L_{v→c} = L_channel(v) + ∑_{c'≠c} R_{c'→v}
+fn variable_node_update(ch_llr: Llr, incoming_c2v: &[Llr], out: &mut [Llr]) {
+    let total: Llr = ch_llr + incoming_c2v.iter().sum::<Llr>();
+    for ((&r, o)) in incoming_c2v.iter().zip(out.iter_mut()) {
+        *o = total - r;
+    }
+}
+
+#[inline]
+fn posterior_llr(ch_llr: Llr, c2v_msgs: &[Llr]) -> Llr {
+    ch_llr + c2v_msgs.iter().sum::<Llr>()
+}
+
+// Messages internes
+// Indexés par (check_id, position_dans_liste_voisins) accès O(1)
+
+struct Messages {
+    v2c: Vec<Vec<Llr>>, // v2c[c][j] : var_neighbor(c)[j] -> check c
+    c2v: Vec<Vec<Llr>>, // c2v[c][j] : check c -> var_neighbor(c)[j]
+}
+
+impl Messages {
+    fn new(graph: &TannerGraph) -> Self {
+        let v2c = (0..graph.n_chk)
+            .map(|c| vec![0.0; graph.chk_degree(c)])
+            .collect();
+        let c2v = (0..graph.n_chk)
+            .map(|c| vec![0.0; graph.chk_degree(c)])
+            .collect();
+        Self { v2c, c2v }
+    }
+}
+
+// Table de correspondance, pour chaque (var, check), index dans la liste de voisins
+// Précalculée une fois à la construction du décodeur
+struct EdgeIndex {
+    // var_pos_in_chk[c][j] : position de var_neighbor(c)[j] dans var_to_chk[v]
+    var_pos_in_chk: Vec<Vec<usize>>,
+    // chk_pos_in_var[v][i] : position de var_neighbor(v)[i] dans chk_to_var[c]
+    chk_pos_in_var: Vec<Vec<usize>>,
+}
+
+impl EdgeIndex {
+    fn build(graph: &TannerGraph) -> Self {
+        let var_pos_in_chk = (0..graph.n_chk)
+            .map(|c| {
+                graph
+                    .chk_neighbors(c)
+                    .iter()
+                    .map(|&v| graph.var_neighbors(v).iter().position(|&x| x == c).unwrap())
+                    .collect()
+            })
+            .collect();
+        let chk_pos_in_var = (0..graph.n_var)
+            .map(|v| {
+                graph
+                    .var_neighbors(v)
+                    .iter()
+                    .map(|&c| graph.chk_neighbors(c).iter().position(|&x| x == v).unwrap())
+                    .collect()
+            })
+            .collect();
+        Self {
+            var_pos_in_chk,
+            chk_pos_in_var,
+        }
+    }
+}
+
+// Bit-Flipping
+
+pub struct BitFlippingDecoder {
+    graph: TannerGraph,
+    h: SparseMatrixGF2,
+    config: DecoderConfig,
+}
+
+impl BitFlippingDecoder {
+    pub fn new(code: &LdpcCode, config: DecoderConfig) -> Self {
+        Self {
+            graph: code.graph.clone(),
+            h: code.h.clone(),
+            config,
+        }
+    }
+}
+
+impl Decoder for BitFlippingDecoder {
+    fn decode(&self, channel_llr: &[Llr]) -> DecoderResult {
+        let mut bits: Vec<Gf2> = channel_llr.iter().map(|&l| hard_decision(l)).collect();
+
+        for _iter in 0..self.config.max_iterations {
+            let syndrome = compute_syndrome(&self.h, &bits);
+            if self.config.early_stopping && syndrome.iter().all(|&s| s == 0) {
+                return DecoderResult::Converged(bits);
+            }
+            let mut unsatisfied = vec![0usize; self.graph.n_var];
+            for c in 0..self.graph.n_chk {
+                if syndrome[c] == 1 {
+                    for &v in self.graph.chk_neighbors(c) {
+                        unsatisfied[v] += 1;
+                    }
+                }
+            }
+            let mut flipped = false;
+            for v in 0..self.graph.n_var {
+                if unsatisfied[v] > self.graph.var_degree(v) / 2 {
+                    bits[v] ^= 1;
+                    flipped = true;
+                }
+            }
+            if !flipped {
+                break;
+            }
+        }
+
+        let synd = compute_syndrome(&self.h, &bits);
+        if synd.iter().all(|&s| s == 0) {
+            DecoderResult::Converged(bits)
+        } else {
+            DecoderResult::MaxIterationsReached(bits)
+        }
+    }
+}
+
+// Noyau BP partagé par SP et MinSum
+
+fn bp_decode<F>(
+    graph: &TannerGraph,
+    h: &SparseMatrixGF2,
+    config: &DecoderConfig,
+    channel_llr: &[Llr],
+    edge_idx: &EdgeIndex,
+    check_update: F,
+) -> DecoderResult
+where
+    F: Fn(&[Llr], &mut [Llr]),
+{
+    let mut msgs = Messages::new(graph);
+
+    // Initialisation : v2c[c][j] = L_channel(var_neighbor(c)[j])
+    for c in 0..graph.n_chk {
+        for (j, &v) in graph.chk_neighbors(c).iter().enumerate() {
+            msgs.v2c[c][j] = channel_llr[v];
+        }
+    }
+
+    for _iter in 0..config.max_iterations {
+        // Mise à jour des check-nodes
+        for c in 0..graph.n_chk {
+            let v2c = msgs.v2c[c].clone();
+            check_update(&v2c, &mut msgs.c2v[c]);
+        }
+
+        // Mise à jour des var-nodes
+        for v in 0..graph.n_var {
+            let neighbors = graph.var_neighbors(v);
+            // Rassembler les c2v entrants sur ce var-node
+            let incoming: Vec<Llr> = neighbors
+                .iter()
+                .enumerate()
+                .map(|(i, &c)| {
+                    let j = edge_idx.chk_pos_in_var[v][i];
+                    msgs.c2v[c][j]
+                })
+                .collect();
+            let mut new_v2c = vec![0.0; neighbors.len()];
+            variable_node_update(channel_llr[v], &incoming, &mut new_v2c);
+            for (i, &c) in neighbors.iter().enumerate() {
+                let j = edge_idx.chk_pos_in_var[v][i];
+                msgs.v2c[c][j] = new_v2c[i];
+            }
+        }
+
+        // Hard décision + arrêt
+        if config.early_stopping {
+            let bits = make_decision(graph, &msgs, channel_llr, edge_idx);
+            if compute_syndrome(h, &bits).iter().all(|&s| s == 0) {
+                return DecoderResult::Converged(bits);
+            }
+        }
+    }
+
+    let bits = make_decision(graph, &msgs, channel_llr, edge_idx);
+    let synd = compute_syndrome(h, &bits);
+    if synd.iter().all(|&s| s == 0) {
+        DecoderResult::Converged(bits)
+    } else {
+        DecoderResult::MaxIterationsReached(bits)
+    }
+}
+
+fn make_decision(
+    graph: &TannerGraph,
+    msgs: &Messages,
+    channel_llr: &[Llr],
+    edge_idx: &EdgeIndex,
+) -> Vec<Gf2> {
+    (0..graph.n_var)
+        .map(|v| {
+            let incoming: Vec<Llr> = graph
+                .var_neighbors(v)
+                .iter()
+                .enumerate()
+                .map(|(i, &c)| {
+                    let j = edge_idx.chk_pos_in_var[v][i];
+                    msgs.c2v[c][j]
+                })
+                .collect();
+            hard_decision(posterior_llr(channel_llr[v], &incoming))
+        })
+        .collect()
+}
+
+// Sum-Product
+
+pub struct SumProductDecoder {
+    graph: TannerGraph,
+    h: SparseMatrixGF2,
+    config: DecoderConfig,
+    edge_idx: EdgeIndex,
+}
+
+impl SumProductDecoder {
+    pub fn new(code: &LdpcCode, config: DecoderConfig) -> Self {
+        let edge_idx = EdgeIndex::build(&code.graph);
+        Self {
+            graph: code.graph.clone(),
+            h: code.h.clone(),
+            config,
+            edge_idx,
+        }
+    }
+}
+
+impl Decoder for SumProductDecoder {
+    fn decode(&self, channel_llr: &[Llr]) -> DecoderResult {
+        bp_decode(
+            &self.graph,
+            &self.h,
+            &self.config,
+            channel_llr,
+            &self.edge_idx,
+            |incoming, out| check_node_update_sp(incoming, out),
+        )
+    }
+}
+
+// Min-Sum
+
+pub struct MinSumDecoder {
+    graph: TannerGraph,
+    h: SparseMatrixGF2,
+    config: DecoderConfig,
+    scaling_factor: Llr,
+    edge_idx: EdgeIndex,
+}
+
+impl MinSumDecoder {
+    pub fn new(code: &LdpcCode, config: DecoderConfig, scaling_factor: Llr) -> Self {
+        let edge_idx = EdgeIndex::build(&code.graph);
+        Self {
+            graph: code.graph.clone(),
+            h: code.h.clone(),
+            config,
+            scaling_factor,
+            edge_idx,
+        }
+    }
+}
+
+impl Decoder for MinSumDecoder {
+    fn decode(&self, channel_llr: &[Llr]) -> DecoderResult {
+        let alpha = self.scaling_factor;
+        bp_decode(
+            &self.graph,
+            &self.h,
+            &self.config,
+            channel_llr,
+            &self.edge_idx,
+            move |incoming, out| check_node_update_ms(incoming, out, alpha),
+        )
+    }
+}
+
+// Factory
+
+#[derive(Debug, Clone)]
+pub enum DecoderMethod {
+    BitFlipping,
+    SumProduct,
+    MinSum { scaling_factor: Llr },
+}
+
+pub fn build_decoder(
+    code: &LdpcCode,
+    method: DecoderMethod,
+    config: DecoderConfig,
+) -> Box<dyn Decoder> {
+    match method {
+        DecoderMethod::BitFlipping => Box::new(BitFlippingDecoder::new(code, config)),
+        DecoderMethod::SumProduct => Box::new(SumProductDecoder::new(code, config)),
+        DecoderMethod::MinSum { scaling_factor } => {
+            Box::new(MinSumDecoder::new(code, config, scaling_factor))
+        }
+    }
+}
+
+// #[cfg(test)]
+// mod tests {
+//     use super::*;
+//
+//     #[test]
+//     fn test_phi_is_involution() {
+//         let x = 2.5_f64;
+//         assert!((phi(phi(x)) - x).abs() < 1e-9);
+//     }
+//
+//     #[test]
+//     fn test_hard_decision() {
+//         assert_eq!(hard_decision(0.1), 0);
+//         assert_eq!(hard_decision(-0.1), 1);
+//         assert_eq!(hard_decision(0.0), 0);
+//     }
+// }
diff --git a/srctmp/encoder.rs b/srctmp/encoder.rs
new file mode 100644
index 0000000..9ea240e
--- /dev/null
+++ b/srctmp/encoder.rs
@@ -0,0 +1,202 @@
+use crate::code::{LdpcCode, SystematicForm};
+use crate::matrix::{DenseMatrixGF2, SparseMatrixGF2};
+use crate::{BitVec, Gf2, LdpcError, Result};
+
+// Trait Encoder
+
+pub trait Encoder: Send + Sync {
+    fn encode(&self, message: &[Gf2]) -> Result<BitVec>;
+    fn message_len(&self) -> usize;
+    fn codeword_len(&self) -> usize;
+
+    fn check_input(&self, msg: &[Gf2]) -> Result<()> {
+        if msg.len() != self.message_len() {
+            return Err(LdpcError::DimensionMismatch {
+                expected: self.message_len(),
+                got: msg.len(),
+            });
+        }
+        Ok(())
+    }
+}
+
+// Méthode d'encodage
+
+#[derive(Debug, Clone)]
+pub enum EncodingMethod {
+    // Via G = [I_k | P]. Complexité O(n^2). Point de départ simple
+    Systematic,
+    // Richardson-Urbanke. Complexité O(n + g^2), g = gap << n
+    RichardsonUrbanke,
+}
+
+// Encodeur systématique
+// c = m * G = [m | m*P]
+// Les bits de parité p = m * P sont calculés par multiplication dense
+// Le mot de code est ensuite réordonné selon la permutation inverse
+
+pub struct SystematicEncoder {
+    k: usize,
+    n: usize,
+    g: DenseMatrixGF2,
+    perm_inv: Vec<usize>,
+}
+
+impl SystematicEncoder {
+    pub fn new(code: &mut LdpcCode) -> Result<Self> {
+        code.compute_systematic_form()?;
+        let sf = code.systematic_form.as_ref().unwrap();
+        Ok(Self {
+            k: code.k(),
+            n: code.n(),
+            g: sf.g.clone(),
+            perm_inv: sf.col_perm_inv.clone(),
+        })
+    }
+}
+
+impl Encoder for SystematicEncoder {
+    fn encode(&self, message: &[Gf2]) -> Result<BitVec> {
+        self.check_input(message)?;
+        let c_perm = self.g.multiply_vec(message);
+        // Réordonner les bits selon la permutation inverse
+        let mut c = vec![0u8; self.n];
+        for (i, &ci) in c_perm.iter().enumerate() {
+            c[self.perm_inv[i]] = ci;
+        }
+        Ok(c)
+    }
+
+    fn message_len(&self) -> usize {
+        self.k
+    }
+    fn codeword_len(&self) -> usize {
+        self.n
+    }
+}
+
+// Encodeur Richardson-Urbanke
+// H est réarrangée par permutations en :
+//
+//      ┌         ┐
+//      │ A  B  T │
+//  H = │         │    T = triangulaire inférieure, φ = -ET^{-1}B + D
+//      │ C  D  E │
+//      └         ┘
+//
+// Encodage en deux étapes :
+//  1. p₂ = φ^{-1} * (-ET^{-1}*As - Cs)       [dense g×g, g = gap]
+//  2. p₁ = T^{-1} * (As + Bp₂)              [substitution arrière]
+// Complexité totale : O(n + g²)
+
+pub struct RichardsonUrbankeEncoder {
+    k: usize,
+    n: usize,
+    a: SparseMatrixGF2,
+    b: SparseMatrixGF2,
+    c: SparseMatrixGF2,
+    d: SparseMatrixGF2,
+    e: SparseMatrixGF2,
+    // T^{-1} : résolution par substitution arrière (T triangulaire inférieure)
+    t_inv: DenseMatrixGF2,
+    // φ^{-1} : petit (g×g), calculé une fois offline
+    phi_inv: DenseMatrixGF2,
+    col_perm_inv: Vec<usize>,
+}
+
+impl RichardsonUrbankeEncoder {
+    pub fn new(code: &LdpcCode) -> Result<Self> {
+        // TODO: implémenter le pré-traitement de H par permutations de lignes/colonnes
+        // pour identifier les blocs A, B, C, D, E, T et calculer phi^{-1}.
+        todo!("Pré-traitement Richardson-Urbanke")
+    }
+
+    // Substitution arrière sur T triangulaire inférieure : résout T*x = b en GF(2)
+    fn backward_substitution(t_inv: &DenseMatrixGF2, b: &[Gf2]) -> Vec<Gf2> {
+        let n = b.len();
+        let mut x = vec![0u8; n];
+        for i in 0..n {
+            let mut s = b[i];
+            for j in 0..i {
+                s ^= t_inv.get(i, j) & x[j];
+            }
+            x[i] = s;
+        }
+        x
+    }
+}
+
+impl Encoder for RichardsonUrbankeEncoder {
+    fn encode(&self, message: &[Gf2]) -> Result<BitVec> {
+        self.check_input(message)?;
+        // Étape 1 : As
+        let a_s = self.a.multiply_vec(message);
+        // Étape 2 : p₂ = φ^{-1} * (E*T^{-1}*As + Cs)
+        let t_inv_as = Self::backward_substitution(&self.t_inv, &a_s);
+        let e_t_inv_as = self.e.multiply_vec(&t_inv_as);
+        let c_s = self.c.multiply_vec(message);
+        let rhs: Vec<Gf2> = e_t_inv_as
+            .iter()
+            .zip(c_s.iter())
+            .map(|(&a, &b)| a ^ b)
+            .collect();
+        let p2 = self.phi_inv.multiply_vec(&rhs);
+        // Étape 3 : p₁ = T^{-1} * (As + Bp₂)
+        let b_p2 = self.b.multiply_vec(&p2);
+        let sum: Vec<Gf2> = a_s.iter().zip(b_p2.iter()).map(|(&a, &b)| a ^ b).collect();
+        let p1 = Self::backward_substitution(&self.t_inv, &sum);
+        // Assemblage et dépermutation
+        let mut c_perm: Vec<Gf2> = message
+            .iter()
+            .chain(p1.iter())
+            .chain(p2.iter())
+            .cloned()
+            .collect();
+        let mut c = vec![0u8; self.n];
+        for (i, &ci) in c_perm.iter().enumerate() {
+            c[self.col_perm_inv[i]] = ci;
+        }
+        Ok(c)
+    }
+
+    fn message_len(&self) -> usize {
+        self.k
+    }
+    fn codeword_len(&self) -> usize {
+        self.n
+    }
+}
+
+// Factory
+
+pub fn build_encoder(code: &mut LdpcCode, method: EncodingMethod) -> Result<Box<dyn Encoder>> {
+    match method {
+        EncodingMethod::Systematic => Ok(Box::new(SystematicEncoder::new(code)?)),
+        EncodingMethod::RichardsonUrbanke => Ok(Box::new(RichardsonUrbankeEncoder::new(code)?)),
+    }
+}
+
+// #[cfg(test)]
+// mod tests {
+//     use super::*;
+//     use crate::code::{CodeTopology, GenerationMethod, LdpcCode, LdpcParams};
+//
+//     fn make_code() -> LdpcCode {
+//         LdpcCode::new(LdpcParams {
+//             n: 12,
+//             k: 6,
+//             topology: CodeTopology::Regular { wc: 2, wr: 4 },
+//             generation: GenerationMethod::Gallager,
+//             seed: Some(42),
+//         })
+//         .unwrap()
+//     }
+//
+//     #[test]
+//     fn test_encoder_wrong_input_length_errors() {
+//         let mut code = make_code();
+//         let enc = SystematicEncoder::new(&mut code).unwrap();
+//         let result = enc.encode(&vec![0u8; 5]); // devrait être 6
+//         assert!(result.is_err());
+//     }
+// }
diff --git a/srctmp/graph.rs b/srctmp/graph.rs
new file mode 100644
index 0000000..6738cfd
--- /dev/null
+++ b/srctmp/graph.rs
@@ -0,0 +1,171 @@
+use crate::matrix::SparseMatrixGF2;
+use std::collections::VecDeque;
+
+// Graphe de Tanner
+
+#[derive(Debug, Clone)]
+pub struct TannerGraph {
+    pub n_var: usize,
+    pub n_chk: usize,
+    var_to_chk: Vec<Vec<usize>>,
+    chk_to_var: Vec<Vec<usize>>,
+}
+
+impl TannerGraph {
+    pub fn from_matrix(h: &SparseMatrixGF2) -> Self {
+        let n_var = h.cols;
+        let n_chk = h.rows;
+        let chk_to_var: Vec<Vec<usize>> = (0..n_chk).map(|c| h.row_neighbors(c).to_vec()).collect();
+        let mut var_to_chk = vec![vec![]; n_var];
+        for c in 0..n_chk {
+            for &v in &chk_to_var[c] {
+                var_to_chk[v].push(c);
+            }
+        }
+        Self {
+            n_var,
+            n_chk,
+            var_to_chk,
+            chk_to_var,
+        }
+    }
+
+    pub fn var_neighbors(&self, v: usize) -> &[usize] {
+        &self.var_to_chk[v]
+    }
+    pub fn chk_neighbors(&self, c: usize) -> &[usize] {
+        &self.chk_to_var[c]
+    }
+    pub fn var_degree(&self, v: usize) -> usize {
+        self.var_to_chk[v].len()
+    }
+    pub fn chk_degree(&self, c: usize) -> usize {
+        self.chk_to_var[c].len()
+    }
+
+    // Calcule le girth par BFS depuis chaque noeud de variable
+    // O(n * (n + m))
+    pub fn girth(&self) -> usize {
+        let mut min_girth = usize::MAX;
+        for start in 0..self.n_var {
+            if let Some(g) = self.bfs_girth_from_var(start) {
+                min_girth = min_girth.min(g);
+                if min_girth == 4 {
+                    return 4;
+                } // minimum possible
+            }
+        }
+        min_girth
+    }
+
+    // Détection rapide de cycles-4, deux var-nodes partagent >= check-nodes
+    pub fn has_4_cycle(&self) -> bool {
+        for v1 in 0..self.n_var {
+            for v2 in (v1 + 1)..self.n_var {
+                let common = self.var_to_chk[v1]
+                    .iter()
+                    .filter(|c| self.var_to_chk[v2].contains(c))
+                    .count();
+                if common >= 2 {
+                    return true;
+                }
+            }
+        }
+        false
+    }
+
+    // Girth local depuis un noeud de variable v (pour PEG).
+    pub fn local_girth_from_var(&self, v: usize) -> usize {
+        self.bfs_girth_from_var(v).unwrap_or(usize::MAX)
+    }
+
+    // BFS depuis le noeud de variable start,
+    // retourne la longueur du court cycle passant par ce noeud (None si aucun cycle)
+    fn bfs_girth_from_var(&self, start: usize) -> Option<usize> {
+        // dist_var[v] = distance depuis start jusqu'au var-node v
+        // dist_chk[c] = distance depuis start jusqu'au check-node c
+        let mut dist_var = vec![usize::MAX; self.n_var];
+        let mut dist_chk = vec![usize::MAX; self.n_chk];
+        dist_var[start] = 0;
+
+        // File : (is_var: bool, index, distance)
+        let mut queue: VecDeque<(bool, usize, usize)> = VecDeque::new();
+        queue.push_back((true, start, 0));
+        let mut shortest = None;
+
+        while let Some((is_var, node, dist)) = queue.pop_front() {
+            if is_var {
+                for &c in self.var_neighbors(node) {
+                    if dist_chk[c] == usize::MAX {
+                        dist_chk[c] = dist + 1;
+                        queue.push_back((false, c, dist + 1));
+                    } else {
+                        // Cycle trouvé
+                        let cycle_len = dist + 1 + dist_chk[c];
+                        shortest = Some(shortest.map_or(cycle_len, |s: usize| s.min(cycle_len)));
+                    }
+                }
+            } else {
+                for &v in self.chk_neighbors(node) {
+                    if v == start && dist + 1 >= 2 {
+                        let cycle_len = dist + 1;
+                        shortest = Some(shortest.map_or(cycle_len, |s: usize| s.min(cycle_len)));
+                        continue;
+                    }
+                    if dist_var[v] == usize::MAX {
+                        dist_var[v] = dist + 1;
+                        queue.push_back((true, v, dist + 1));
+                    }
+                }
+            }
+        }
+        shortest
+    }
+
+    pub fn var_degree_distribution(&self) -> Vec<f64> {
+        let max_deg = self.var_to_chk.iter().map(|v| v.len()).max().unwrap_or(0);
+        let mut counts = vec![0usize; max_deg + 1];
+        for v in 0..self.n_var {
+            counts[self.var_degree(v)] += 1;
+        }
+        counts
+            .iter()
+            .map(|&c| c as f64 / self.n_var as f64)
+            .collect()
+    }
+
+    pub fn is_regular(&self) -> bool {
+        let d0 = self.var_degree(0);
+        let c0 = self.chk_degree(0);
+        self.var_to_chk.iter().all(|v| v.len() == d0)
+            && self.chk_to_var.iter().all(|c| c.len() == c0)
+    }
+}
+
+// #[cfg(test)]
+// mod tests {
+//     use super::*;
+//     use crate::matrix::SparseMatrixGF2;
+//
+//     fn simple_h() -> SparseMatrixGF2 {
+//         SparseMatrixGF2::from_dense(&vec![
+//             vec![1u8, 1, 0, 1, 0],
+//             vec![0, 1, 1, 0, 1],
+//             vec![1, 0, 1, 0, 1],
+//         ])
+//     }
+//
+//     #[test]
+//     fn test_construction_from_matrix() {
+//         let h = simple_h();
+//         let g = TannerGraph::from_matrix(&h);
+//         assert_eq!(g.n_var, 5);
+//         assert_eq!(g.n_chk, 3);
+//     }
+//
+//     #[test]
+//     fn test_var_degree() {
+//         let g = TannerGraph::from_matrix(&simple_h());
+//         assert_eq!(g.var_degree(0), 2);
+//     }
+// }
diff --git a/srctmp/lib.rs b/srctmp/lib.rs
new file mode 100644
index 0000000..6dca682
--- /dev/null
+++ b/srctmp/lib.rs
@@ -0,0 +1,38 @@
+pub mod channel;
+pub mod code;
+pub mod decoder;
+pub mod encoder;
+pub mod graph;
+pub mod matrix;
+
+pub type Gf2 = u8;
+pub type Llr = f64;
+pub type BitVec = Vec<Gf2>;
+
+#[derive(Debug, thiserror::Error)]
+pub enum LdpcError {
+    #[error("Paramètres invalides : {0}")]
+    InvalidParameters(String),
+
+    #[error("Matrice singulière : impossible d'inverser")]
+    SingularMatrix,
+
+    #[error("Génération échouée après {attempts} tentatives")]
+    GenerationFailed { attempts: usize },
+
+    #[error("Dimension incorrecte : attendu {expected}, reçu {got}")]
+    DimensionMismatch { expected: usize, got: usize },
+
+    #[error("Le vecteur fourni n'est pas un mot de code valide")]
+    InvalidCodeword,
+
+    #[error("Paramètre hors plage : {0}")]
+    OutOfRange(String),
+}
+
+pub type Result<T> = std::result::Result<T, LdpcError>;
+
+pub use channel::Channel;
+pub use code::{CodeTopology, GenerationMethod, LdpcCode, LdpcParams};
+pub use decoder::{Decoder, DecoderConfig, DecoderMethod, DecoderResult};
+pub use encoder::{Encoder, EncodingMethod};
diff --git a/srctmp/main.rs b/srctmp/main.rs
new file mode 100644
index 0000000..490b7f9
--- /dev/null
+++ b/srctmp/main.rs
@@ -0,0 +1,60 @@
+use ldpc::{
+    channel::{AwgnChannel, Channel},
+    code::{CodeTopology, GenerationMethod, LdpcCode, LdpcParams},
+    decoder::{build_decoder, DecoderConfig, DecoderMethod},
+    encoder::{build_encoder, EncodingMethod},
+};
+use rand::{Rng, SeedableRng};
+
+fn main() -> ldpc::Result<()> {
+    // ── 1. Générer le code ────────────────────────────────────────────────────
+    let params = LdpcParams {
+        n: 100,
+        k: 50,
+        topology: CodeTopology::Regular { wc: 3, wr: 6 },
+        generation: GenerationMethod::MacKayNeal { max_attempts: 500 },
+        seed: Some(42),
+    };
+    let mut code = LdpcCode::new(params)?;
+    println!(
+        "Code LDPC : n={}, k={}, taux={:.2}, girth={}",
+        code.n(),
+        code.k(),
+        code.rate(),
+        code.girth()
+    );
+
+    // ── 2. Encodeur ───────────────────────────────────────────────────────────
+    let encoder = build_encoder(&mut code, EncodingMethod::Systematic)?;
+
+    // ── 3. Canal ──────────────────────────────────────────────────────────────
+    let channel = AwgnChannel::new(2.0, code.rate())?; // 2 dB Eb/N0
+    println!("Capacité AWGN ≈ {:.4} bits/utilisation", channel.capacity());
+
+    // ── 4. Décodeur ───────────────────────────────────────────────────────────
+    let decoder = build_decoder(&code, DecoderMethod::SumProduct, DecoderConfig::default());
+
+    // ── 5. Simulation ─────────────────────────────────────────────────────────
+    let mut rng = rand::rngs::StdRng::seed_from_u64(123);
+    let n_trials = 100;
+    let mut errors = 0usize;
+
+    for _ in 0..n_trials {
+        // Message aléatoire
+        let message: Vec<u8> = (0..code.k()).map(|_| rng.gen::<u8>() & 1).collect();
+        // Encodage
+        let codeword = encoder.encode(&message)?;
+        // Transmission AWGN
+        let received_llr = channel.transmit(&codeword, &mut rng);
+        // Décodage
+        let result = decoder.decode(&received_llr);
+        // Vérification
+        if !result.is_success() {
+            errors += 1;
+        }
+    }
+
+    let ber = errors as f64 / n_trials as f64;
+    println!("FER sur {} essais à 2dB : {:.2}%", n_trials, ber * 100.0);
+    Ok(())
+}
diff --git a/srctmp/matrix.rs b/srctmp/matrix.rs
new file mode 100644
index 0000000..9dd1f7a
--- /dev/null
+++ b/srctmp/matrix.rs
@@ -0,0 +1,295 @@
+use crate::{Gf2, Result};
+
+// Matrice creuse GF(2) — format CSR + CSC dual
+
+#[derive(Debug, Clone)]
+pub struct SparseMatrixGF2 {
+    pub rows: usize,
+    pub cols: usize,
+    // CSR — accès ligne i : col_idx[ row_ptr[i] .. row_ptr[i+1] ]
+    row_ptr: Vec<usize>,
+    col_idx: Vec<usize>,
+    // CSC — accès col j  : row_idx[ col_ptr[j] .. col_ptr[j+1] ]
+    col_ptr: Vec<usize>,
+    row_idx: Vec<usize>,
+}
+
+impl SparseMatrixGF2 {
+    pub fn zeros(rows: usize, cols: usize) -> Self {
+        Self {
+            rows,
+            cols,
+            row_ptr: vec![0; rows + 1],
+            col_idx: vec![],
+            col_ptr: vec![0; cols + 1],
+            row_idx: vec![],
+        }
+    }
+
+    // Depuis une liste de (row, col) indiquant les positions des 1s.
+    // Trie les entrées et construit CSR + CSC en un seul passage.
+    pub fn from_positions(rows: usize, cols: usize, mut ones: Vec<(usize, usize)>) -> Self {
+        // Construction CSR
+        ones.sort_unstable();
+        let mut row_ptr = vec![0usize; rows + 1];
+        let mut col_idx = Vec::with_capacity(ones.len());
+        for &(r, c) in &ones {
+            row_ptr[r + 1] += 1;
+            col_idx.push(c);
+        }
+        for i in 0..rows {
+            row_ptr[i + 1] += row_ptr[i];
+        }
+        // Construction CSC
+        let mut col_sorted = ones.clone();
+        col_sorted.sort_unstable_by_key(|&(r, c)| (c, r));
+        let mut col_ptr = vec![0usize; cols + 1];
+        let mut row_idx = Vec::with_capacity(ones.len());
+        for &(r, c) in &col_sorted {
+            col_ptr[c + 1] += 1;
+            row_idx.push(r);
+        }
+        for j in 0..cols {
+            col_ptr[j + 1] += col_ptr[j];
+        }
+        Self {
+            rows,
+            cols,
+            row_ptr,
+            col_idx,
+            col_ptr,
+            row_idx,
+        }
+    }
+
+    pub fn from_dense(dense: &[Vec<Gf2>]) -> Self {
+        let rows = dense.len();
+        let cols = if rows > 0 { dense[0].len() } else { 0 };
+        let ones: Vec<(usize, usize)> = dense
+            .iter()
+            .enumerate()
+            .flat_map(|(r, row)| {
+                row.iter()
+                    .enumerate()
+                    .filter(|(_, &v)| v == 1)
+                    .map(move |(c, _)| (r, c))
+            })
+            .collect();
+        Self::from_positions(rows, cols, ones)
+    }
+
+    pub fn get(&self, row: usize, col: usize) -> Gf2 {
+        let slice = self.row_neighbors(row);
+        if slice.binary_search(&col).is_ok() {
+            1
+        } else {
+            0
+        }
+    }
+
+    // Indices des colonnes où la ligne `row` vaut 1 (voisins check→var)
+    pub fn row_neighbors(&self, row: usize) -> &[usize] {
+        &self.col_idx[self.row_ptr[row]..self.row_ptr[row + 1]]
+    }
+
+    // Indices des lignes où la colonne `col` vaut 1 (voisins var→check)
+    pub fn col_neighbors(&self, col: usize) -> &[usize] {
+        &self.row_idx[self.col_ptr[col]..self.col_ptr[col + 1]]
+    }
+
+    pub fn row_weight(&self, row: usize) -> usize {
+        self.row_ptr[row + 1] - self.row_ptr[row]
+    }
+
+    pub fn col_weight(&self, col: usize) -> usize {
+        self.col_ptr[col + 1] - self.col_ptr[col]
+    }
+
+    pub fn nnz(&self) -> usize {
+        self.col_idx.len()
+    }
+
+    pub fn density(&self) -> f64 {
+        self.nnz() as f64 / (self.rows * self.cols) as f64
+    }
+
+    // Produit H * x mod 2 (calcul du syndrome : s = H * c^T)
+    pub fn multiply_vec(&self, x: &[Gf2]) -> Vec<Gf2> {
+        (0..self.rows)
+            .map(|r| {
+                self.row_neighbors(r)
+                    .iter()
+                    .map(|&c| x[c])
+                    .fold(0u8, |acc, b| acc ^ b)
+            })
+            .collect()
+    }
+
+    pub fn transpose(&self) -> Self {
+        Self {
+            rows: self.cols,
+            cols: self.rows,
+            row_ptr: self.col_ptr.clone(),
+            col_idx: self.row_idx.clone(),
+            col_ptr: self.row_ptr.clone(),
+            row_idx: self.col_idx.clone(),
+        }
+    }
+
+    // Vérifie si deux colonnes partagent >= 2 positions de 1 -> cycle-4 détecté
+    pub fn columns_share_two_ones(&self, c1: usize, c2: usize) -> bool {
+        let n1 = self.col_neighbors(c1);
+        let n2 = self.col_neighbors(c2);
+        let mut common = 0usize;
+        let (mut i, mut j) = (0, 0);
+        while i < n1.len() && j < n2.len() {
+            match n1[i].cmp(&n2[j]) {
+                std::cmp::Ordering::Less => i += 1,
+                std::cmp::Ordering::Greater => j += 1,
+                std::cmp::Ordering::Equal => {
+                    common += 1;
+                    if common >= 2 {
+                        return true;
+                    }
+                    i += 1;
+                    j += 1;
+                }
+            }
+        }
+        false
+    }
+
+    pub fn to_dense(&self) -> Vec<Vec<Gf2>> {
+        let mut out = vec![vec![0u8; self.cols]; self.rows];
+        for r in 0..self.rows {
+            for &c in self.row_neighbors(r) {
+                out[r][c] = 1;
+            }
+        }
+        out
+    }
+}
+
+// Matrice dense GF(2)
+// Utilisée pour la matrice génératrice G et les calculs de Gauss-Jordan
+// G = [I | P], P est souvent dense
+
+#[derive(Debug, Clone)]
+pub struct DenseMatrixGF2 {
+    pub rows: usize,
+    pub cols: usize,
+    data: Vec<Vec<Gf2>>,
+}
+
+impl DenseMatrixGF2 {
+    pub fn zeros(rows: usize, cols: usize) -> Self {
+        Self {
+            rows,
+            cols,
+            data: vec![vec![0u8; cols]; rows],
+        }
+    }
+
+    pub fn identity(n: usize) -> Self {
+        let mut m = Self::zeros(n, n);
+        for i in 0..n {
+            m.data[i][i] = 1;
+        }
+        m
+    }
+
+    pub fn get(&self, row: usize, col: usize) -> Gf2 {
+        self.data[row][col]
+    }
+    pub fn set(&mut self, row: usize, col: usize, val: Gf2) {
+        self.data[row][col] = val;
+    }
+
+    // Addition de deux lignes dans GF(2)
+    pub fn row_add(&mut self, dst: usize, src: usize) {
+        for j in 0..self.cols {
+            self.data[dst][j] ^= self.data[src][j];
+        }
+    }
+
+    pub fn row_swap(&mut self, r1: usize, r2: usize) {
+        self.data.swap(r1, r2);
+    }
+
+    pub fn multiply_vec(&self, x: &[Gf2]) -> Vec<Gf2> {
+        self.data
+            .iter()
+            .map(|row| {
+                row.iter()
+                    .zip(x.iter())
+                    .fold(0u8, |acc, (&a, &b)| acc ^ (a & b))
+            })
+            .collect()
+    }
+
+    pub fn into_sparse(self) -> SparseMatrixGF2 {
+        SparseMatrixGF2::from_dense(&self.data)
+    }
+
+    // Retourne la permutation de colonnes appliquée et le rang
+    // self est sous forme échelonnée réduite
+    pub fn gauss_jordan_gf2(&mut self) -> (Vec<usize>, usize) {
+        let mut perm: Vec<usize> = (0..self.cols).collect();
+        let mut pivot_row = 0;
+        for col in 0..self.cols {
+            // Chercher un pivot dans la colonne courante
+            let pivot = (pivot_row..self.rows).find(|&r| self.data[r][col] == 1);
+            let Some(p) = pivot else { continue };
+            self.row_swap(pivot_row, p);
+            // Éliminer dans toutes les autres lignes
+            for r in 0..self.rows {
+                if r != pivot_row && self.data[r][col] == 1 {
+                    self.row_add(r, pivot_row);
+                }
+            }
+            pivot_row += 1;
+            if pivot_row == self.rows {
+                break;
+            }
+        }
+        (perm, pivot_row)
+    }
+}
+
+// #[cfg(test)]
+// mod tests {
+//     use super::*;
+//
+//     #[test]
+//     fn test_from_dense_roundtrip() {
+//         let dense = vec![vec![1u8, 0, 1, 0], vec![0, 1, 0, 1], vec![1, 1, 0, 0]];
+//         let sparse = SparseMatrixGF2::from_dense(&dense);
+//         assert_eq!(sparse.to_dense(), dense);
+//     }
+//
+//     #[test]
+//     fn test_multiply_vec_gf2() {
+//         let dense = vec![vec![1u8, 1, 0], vec![0, 1, 1]];
+//         let h = SparseMatrixGF2::from_dense(&dense);
+//         let x = vec![1u8, 1, 1];
+//         let s = h.multiply_vec(&x);
+//         // ligne 0 : 1^1^0 = 0, ligne 1 : 0^1^1 = 0
+//         assert_eq!(s, vec![0u8, 0]);
+//     }
+//
+//     #[test]
+//     fn test_transpose_double_is_identity() {
+//         let dense = vec![vec![1u8, 0, 1], vec![0, 1, 0]];
+//         let h = SparseMatrixGF2::from_dense(&dense);
+//         assert_eq!(h.transpose().transpose().to_dense(), dense);
+//     }
+//
+//     #[test]
+//     fn test_columns_share_two_ones() {
+//         // Colonnes 0 et 1 partagent les lignes 0 et 1 → cycle-4
+//         let dense = vec![vec![1u8, 1, 0], vec![1, 1, 0], vec![0, 0, 1]];
+//         let h = SparseMatrixGF2::from_dense(&dense);
+//         assert!(h.columns_share_two_ones(0, 1));
+//         assert!(!h.columns_share_two_ones(0, 2));
+//     }
+// }