Concentric Permutation Source Codes

Permutation codes are a class of structured vector quantizers with a computationally-simple encoding procedure based on sorting the scalar components. Using a codebook comprising several permutation codes as subcodes preserves the simplicity of encoding while increasing the number of rate-distortion operating points, improving the convex hull of operating points, and increasing design complexity. We show that when the subcodes are designed with the same composition, optimization of the codebook reduces to a lower-dimensional vector quantizer design within a single cone. Heuristics for reducing design complexity are presented, including an optimization of the rate allocation in a shape-gain vector quantizer with gain-dependent wrapped spherical shape codebook.

💡 Research Summary

The paper introduces “Concentric Permutation Codes” (CPC), an extension of classical permutation source codes (PSCs) that places codewords on several concentric spheres rather than a single sphere. Classical PSCs achieve very low encoding complexity (O(n log n)) by sorting the input vector and mapping its ordered components to the ordered entries of a single initial codeword; all codewords are generated by permuting that initial vector. While this structure is attractive, a single‑sphere placement limits the achievable rate–distortion performance, especially at moderate rates, because the codebook cannot fully exploit the available Euclidean space.

CPC resolves this limitation by allowing J different initial codewords, each defining its own permutation sub‑codebook. The union of these sub‑codebooks forms a codebook that lives on J concentric spheres. Encoding proceeds in two steps: (1) sort the input, generate the nearest codeword in each sub‑codebook (still O(n log n) plus O(J n) re‑ordering), and (2) select the closest among the J candidates (O(J n)). As long as J grows at most logarithmically with n, the overall complexity remains comparable to a single‑sphere PSC.

Designing CPC is considerably more challenging than designing a PSC because one must choose, for each sub‑codebook, both the composition (the integer partition of n that determines how many times each scalar value repeats) and the scalar values themselves. Exhaustive search over all compositions quickly becomes intractable. The authors therefore propose two complementary strategies to reduce this complexity.

First, they consider the special case where all sub‑codebooks share a common composition (n₁,…,n_K). Under this restriction the distortion can be expressed as
D = (1/n) E