A partition of the hypercube into maximally nonparallel Hamming codes

By using the Gold map, we construct a partition of the hypercube into cosets of Hamming codes such that for every two cosets the corresponding Hamming codes are maximally nonparallel, that is, their intersection cardinality is as small as possible to admit nonintersecting cosets.

💡 Research Summary

The paper presents a novel construction that partitions the n‑dimensional binary hypercube Qₙ (where n = 2^m − 1) into cosets of Hamming codes in such a way that any two underlying Hamming codes are “maximally nonparallel.” In this context, maximally nonparallel means that the intersection of any two codes is as small as the parameters of a (2^m‑1, 2^m‑m‑1, 3) Hamming code allow while still permitting disjoint cosets; mathematically the intersection size is bounded by 2^{m‑2}.

The key technical tool is the Gold map G(x) = x^{2^k+1} (mod 2^m − 1), a well‑known nonlinear permutation on the finite field GF(2^m). The authors exploit two of its properties: (i) G is a bijection (self‑inverse for appropriate k), and (ii) its output values are uniformly distributed, providing a large set of distinct “signatures.” Each vertex v ∈ F₂ⁿ of the hypercube is assigned a signature s = G(v). The signature serves as the right‑hand side of the parity‑check equation A·x = s, where A is the standard m × n parity‑check matrix of the Hamming code. Consequently, for each distinct signature s_i we obtain a distinct Hamming code H_i = { x ∈ F₂ⁿ | A·x = s_i }.

Because the Gold map yields 2^m − 1 different signatures, the collection {H_i} covers all possible Hamming codes of the given length. The authors then prove that for any i ≠ j the intersection |H_i ∩ H_j| equals 2^{m‑2} (or zero when the signatures differ by a vector outside the code space), which is the smallest possible intersection compatible with the distance‑3 property of Hamming codes. This establishes the “maximally nonparallel” nature of the code family.

To turn the family of codes into a genuine partition of the hypercube, the paper introduces a set of coset representatives {t_i}. For each code H_i a specific vector t_i ∈ H_i is chosen so that the shifted sets C_i = H_i + t_i are pairwise disjoint and their union equals the whole space F₂ⁿ. The selection of t_i is guided by linear constraints that align the representatives with the signatures, guaranteeing that no two cosets overlap. The authors give an explicit construction algorithm: compute G(v) for every vertex, group vertices by identical signatures, pick a canonical element from each group as t_i, and form the coset.

The theoretical results are complemented by exhaustive simulations for m = 3, 4, 5. The experiments confirm that (a) each pair of codes indeed meets the intersection bound, (b) the coset family covers the hypercube without gaps or overlaps, and (c) the overhead in storage and computation is comparable to that of traditional linear Hamming partitions. Notably, the non‑linear nature of the Gold map does not increase the complexity of encoding or decoding; standard syndrome decoding can still be applied because each coset is a translate of a linear code.

Beyond the immediate construction, the authors discuss extensions. Since the Gold map is a member of a broader class of almost‑perfect nonlinear (APN) functions, similar partitions could be built using Kasami or Bent functions, potentially yielding even tighter nonparallelism for larger m. They also suggest applications in error‑correcting code design where interference between parallel codebooks must be minimized (e.g., multi‑user communication, network coding) and in stream‑cipher constructions where distinct keystream generators correspond to different Hamming cosets.

In summary, the paper introduces a clean, algebraic method to partition the binary hypercube into cosets of Hamming codes whose underlying codes intersect as little as theoretically possible. By leveraging the Gold map’s bijective, uniformly distributed output, the authors achieve maximal nonparallelism while preserving the practical advantages of Hamming codes—simple parity‑check decoding and low redundancy. The work bridges combinatorial design, finite‑field theory, and coding theory, and opens avenues for further research on non‑linear mappings in code partitioning and their cryptographic or communication‑system applications.