diff --git a/src/core/functions.h b/src/core/functions.h
index bb9c03a..44010be 100644
--- a/src/core/functions.h
+++ b/src/core/functions.h
@@ -13,6 +13,7 @@
 #include <variant>
 
 #include "gf2.h"
+#include "sparse.h"
 
 #define TIME_PROF(log, exec, unit)                                                                                                             \
     do                                                                                                                                         \
diff --git a/src/core/sparse.h b/src/core/sparse.h
new file mode 100644
index 0000000..f2e9bbb
--- /dev/null
+++ b/src/core/sparse.h
@@ -0,0 +1,147 @@
+#pragma once
+
+#include <vector>
+#include <iostream>
+#include <sstream>
+
+namespace ldpc
+{
+    template <typename T>
+    struct edge
+    {
+        int rowIndex;
+        int colIndex;
+        T value;
+    };
+
+    struct node
+    {
+        int nodeIndex;
+        int edgeIndex;
+    };
+
+    /**
+     * @brief Sparse matrix with minimal functionality required for encoding/decoding
+     * LDPC codes and other linear block codes.
+     * 
+     * @tparam T finite field
+     */
+    template <typename T>
+    class sparse_csr
+    {
+    public:
+        sparse_csr() = default;
+        sparse_csr(const int m, const int n)
+            : numCols(n),
+              numRows(m),
+              colN(numCols, std::vector<node>()),
+              rowN(numRows, std::vector<node>())
+        {
+        }
+
+        void read_from_file(const std::string &filename, int skipLines);
+
+        std::vector<T> multiply_left(const std::vector<T> &left);   // for encoding
+        std::vector<T> multiply_right(const std::vector<T> &right); // for syndrome
+
+        const int num_cols() const { return numCols; }
+        const int num_rows() const { return numRows; }
+        const std::vector<std::vector<node>> &col_neighbor() const { return colN; }
+        const std::vector<std::vector<node>> &row_neighbor() const { return rowN; }
+        const std::vector<edge<T>> &nz_entry() const { return nonZeroVals; }
+
+    private:
+        int numCols;                         // number of columns
+        int numRows;                         // number of rows
+        std::vector<std::vector<node>> colN; // column neighbors, with row and edge index
+        std::vector<std::vector<node>> rowN; // row neigbors, with col and edge index
+        std::vector<edge<T>> nonZeroVals;    // edges, i.e. non-zero entries with row and col index
+    };
+
+    template <typename T>
+    void sparse_csr<T>::read_from_file(const std::string &filename, int skipLines)
+    {
+        int index = 0;
+
+        std::ifstream infile(filename);
+        std::string line;
+
+        // skip lines at beginning of file
+        while (skipLines-- > 0)
+        {
+            std::getline(infile, line);
+        }
+
+        while (getline(infile, line))
+        {
+            edge<T> entry;
+
+            std::istringstream record(line);
+            record >> entry.rowIndex;
+            record >> entry.colIndex;
+
+            // no value is given
+            if (record)
+            {
+                record >> entry.value;
+            }
+            else
+            {
+                entry.value = T(1);
+            }
+
+            nonZeroVals.push_back(entry);
+
+            colN[entry.colIndex].push_back(node({entry.rowIndex, index}));
+            rowN[entry.rowIndex].push_back(node({entry.colIndex, index}));
+
+            ++index;
+        }
+    }
+
+    /**
+     * @brief Multiply vector from left handside with matrix over field T
+     * 
+     * @tparam T finite field
+     * @param left Row vector
+     * @return std::vector<T> 
+     */
+    template <typename T>
+    std::vector<T> sparse_csr<T>::multiply_left(const std::vector<T> &left)
+    {
+        std::vector<T> result(numCols);
+
+        for (int j = 0; j < numCols; ++j)
+        {
+            for (const auto &n : colN[j])
+            {
+                result[j] += left[n.nodeIndex] * nonZeroVals[n.edgeIndex].value;
+            }
+        }
+
+        return result;
+    }
+
+    /**
+     * @brief Multiply vector from right handside with matrix over field T
+     * 
+     * @tparam T finite field
+     * @param right Column vector
+     * @return std::vector<T> 
+     */
+    template <typename T>
+    std::vector<T> sparse_csr<T>::multiply_right(const std::vector<T> &right)
+    {
+        std::vector<T> result(numRows);
+
+        for (int i = 0; i < numRows; ++i)
+        {
+            for (const auto &n : rowN[i])
+            {
+                result[i] += right[n.nodeIndex] * nonZeroVals[n.edgeIndex].value;
+            }
+        }
+
+        return result;
+    }
+} // namespace ldpc
diff --git a/src/sim_cpu.cpp b/src/sim_cpu.cpp
index 504ee9e..43f3887 100644
--- a/src/sim_cpu.cpp
+++ b/src/sim_cpu.cpp
@@ -5,7 +5,7 @@
 int main(int argc, char *argv[])
 {
     argparse::ArgumentParser parser("ldpc_sim");
-    parser.add_argument("codefile").help("LDPC codefile containing all non-zero entries, column major ordering.");
+    parser.add_argument("codefile").help("LDPC codefile containing all non-zero entries, compressed sparse row (CSR) format.");
     parser.add_argument("output-file").help("Results output file.");
     parser.add_argument("snr-range").help("{MIN} {MAX} {STEP}").nargs(3).action([](const std::string &s) { return std::stod(s); });