Forney's Cubing Construction of Golay Codes

• Forney’s cubing construction is a structured method that generates the optimal (24,12,8) Golay code using nested binary codes with triple redundancy.
• The construction leverages explicit generator matrices from single-parity-check, Reed–Muller, and repetition codes combined via a cubing operator to form a 24-bit codeword.
• Modern adaptations integrate PAC-based decoding, enabling efficient, parallel decoding of the Golay code and related lattices such as the Leech lattice.

Forney's cubing construction provides a structured, algebraic method for generating the binary extended Golay code, an optimal $(24,12,8)$ linear code. This technique, first introduced by Forney in "Coset Codes II" (1988), assembles the length-24 codeword by composing three shorter, nested binary block codes of length 8, using a specific “cubing” operator. Recent work, notably Ji et al. (Ji et al., 2 Feb 2026), has revisited and refined this structural approach, demonstrating its connection to modern Polarization Adjusted Convolutional (PAC) codes and leveraging it for efficient decoding and lattice construction.

1. Overview of the Cubing Construction

Forney’s cubing construction synthesizes the $(24,12,8)$ extended Golay code by stacking three nested binary codes:

• $C_1 \supset C_2 \supset C_3$ on $\mathbb F_2^8$
• $C_1$: $(8,7,2)$ single-parity-check code
• $C_2$: $(8,4,4)$ code, specifically the first-order Reed–Muller code $\operatorname{RM}(1,3)$
• $C_3$: $(8,1,8)$ repetition code

A codeword is constructed as $(x, y, z) = (a + c,\, b + c,\, c)$ for $a \in C_1$, $b \in C_2$, and $c \in C_3$. This “cubing” creates a $24$-bit word, generating a $(24,12,8)$ code with the Golay parameters.

2. Generator Matrices for Component Codes

The construction depends on explicit generator matrices:

• $C_1$:

$G^*(8,7) = \left[I_7\,|\,\mathbf{1}_7\right]$

where $I_7$ is the $7 \times 7$ identity matrix, and $\mathbf{1}_7$ is a column vector of ones.

• $C_2$ (Reed–Muller $\operatorname{RM}(1,3)$): Begin with $F_2^{\otimes 3}$ and select the four rows of weight $\geq 4$ to get

$$G_8 = \begin{bmatrix} 1&1&1&1&0&0&0&0\ 1&1&0&0&1&1&0&0\ 1&0&1&0&1&0&1&0\ 1&1&1&1&1&1&1&1 \end{bmatrix}$$

• $C_2'$: A column-permutation $\pi_3 = [5, 4, 2, 3, 1, 6, 7, 8]$ of $C_2$ gives

$$G_8' = \begin{bmatrix} 0&1&1&1&1&0&0&0\ 1&0&0&1&1&1&0&0\ 1&1&0&0&1&0&1&0\ 1&1&1&1&1&1&1&1 \end{bmatrix}$$

• $C_3$: The all-ones $(8,1,8)$ repetition code:

$G_{\mathrm{rep}} = [1\,1\,1\,1\,1\,1\,1\,1]$

3. Algebraic Assembly of the Golay Generator

The full $(24,12,8)$ code is constructed via Kronecker products and block assembly. Let $S$ and $R$ be:

$$S = \begin{bmatrix} 1 & 0 & 1 \ 0 & 1 & 1 \end{bmatrix}, \quad R = [1\,1\,1]$$

where $S$ generates the $(3,2,2)$ SPC code and $R$ the $(3,1,3)$ repetition code.

The large generator matrix is:

$\hat{G} = \begin{bmatrix} S \otimes G_8 \ R \otimes G_8' \end{bmatrix} = \begin{bmatrix} G_8 & 0 & G_8 \ 0 & G_8 & G_8 \ \hline G_8' & G_8' & G_8' \end{bmatrix}$

The top two $4 \times 24$ blocks (from $S\otimes G_8$) correspond to the indices of $C_2$ and $C_1$; the bottom block (from $R\otimes G_8'$) encodes the “cube”-symmetry from $C_3$.

4. The Cubing Operator: Definition and Clarification

The cubing operator $\square$ is formally defined for nested codes $C_1 \supset C_2 \supset C_3 \subset \mathbb F_2^m$ as:

$$\square(a, b, c) = (a + c,\, b + c,\, c) \in \mathbb F_2^{3m}$$

with $a \in C_1$, $b \in C_2$, $c \in C_3$.

For $m=8$ (the Golay code case), this process precisely generates the $(24,12,8)$ code from the described component codes. The role of $C_3$ in each segment provides structural redundancy (triple coverage), with the additional degrees of freedom assigned via $C_2$ and $C_1$.

5. Specific Parameter Choices in Construction

Parameter selection in Ji et al. aligns with the standard Golay instantiation:

Code	Parameters	Generator Matrix	Notes
$C_1$	$(8,7,2)$	$G^*(8,7) = [I_7\,|\,\mathbf{1}_7]$	Single-parity-check
$C_2$	$(8,4,4)$	$G_8$ as from $\operatorname{RM}(1,3)$	Reed–Muller, four heavy rows
$C_2'$	$(8,4,4)$	$G_8'$ (column permutation $\pi_3$)	Symmetric structural redundancy
$C_3$	$(8,1,8)$	$G_{\mathrm{rep}}$ (all-ones)	Repetition

This explicit structure facilitates efficient encoding and algebraic manipulation.

6. Integration into Modern PAC-Based Decoding

Building on the cubing construction, Ji et al. (Ji et al., 2 Feb 2026) identify a direct connection to the polar generator $G_{24}^p = F_2^{\otimes 3} \otimes F_2^{\otimes 3}$ (a $3 \times 3'$ polar “cube”). They demonstrate that, for suitable sets and pre-transforms,

$$\left[v_\mathcal{A},\,0_{12}\right]T G_{24}^p = v_\mathcal{A} \hat{G}$$

where $\mathcal{A} \subset [24]$, $|\mathcal{A}| = 12$, and $T$ is an upper-triangular convolutional transform.

Three distinct $3 \times 3$ polar kernels $F_3^{(i)}$ and pre-transforms $T_i$ give rise to three PAC code representations of the Golay code. This enables parallel decoding: three SCL (successive cancellation list) decoders, each with list size $L$, yield $3L$ codeword candidates, allowing selection of the most likely path. Empirical results show near-ML decoding is attained with $L\approx 8$, and this method obviates ad hoc column permutations or codeword puncturing.

The approach generalizes to parallel decoding of related lattices such as the Leech lattice $\Lambda_{24}$ and its sublattice $H_{24}$.

7. Broader Significance and Applications

Forney’s cubing construction, especially as modernized through PAC-based techniques, provides:

• Structural insight into the Golay code’s deep symmetry and triple redundancy
• Algebraic pathways for efficient implementation and algebraic enumeration of codewords
• Foundations for small-list parallel list decoders achieving near-ML performance without manual index manipulation
• Transferability to the construction and efficient decoding of high-dimensional lattices, such as $H_{24}$ and $\Lambda_{24}$

A plausible implication is that these canonical algebraic decompositions may further inform code design and application in multilevel lattice decoders and other domains seeking both algebraic regularity and decoding efficiency (Ji et al., 2 Feb 2026).