Extending the known families of scalable Huffman sequences

TL;DR

This paper introduces new constructions of scalable Huffman sequences with delta-like auto-correlations, extending known families and providing practical methods for deconvolving data in multidimensional applications.

Contribution

It presents novel methods for constructing canonical Huffman sequences of various lengths, including complex-scaled and Fibonacci-based arrays, and introduces a two-mask de-correlation technique for imaging.

Findings

New Huffman sequences of lengths 4n+1, 2n, and arbitrary lengths are constructed.

Fibonacci-based arrays with perfect auto-correlation properties are developed.

A two-mask de-correlation method reduces radiation dose in imaging applications.

Abstract

A canonical Huffman sequence is characterized by a zero inner-product between itself and each of its shifted copies, except at their largest relative shifts: their aperiodic auto-correlation then becomes delta-like, a single central peak surrounded by zeros, with one non-zero entry at each end. Prior work showed that the few known families of Huffman sequences (of length $N = 4n-1$, for integers $n > 1$, with continuously scalable elements) are based upon Fibonacci polynomials. Related multi-dimensional ($nD$) Huffman arrays were designed, as well as non-canonical quasi-Huffman arrays that also possess delta-like auto-correlations. We examined links between these discrete sequences and delta-correlated functions defined on the continuum, and provided simple non-iterative approaches to successfully deconvolve $nD$ data blurred by diffuse Huffman arrays. Here we describe new constructions for canonical Huffman sequences. Examples of length $N = 4n+1$, $N = 2n$ and families of arbitrary length are given, including scaled forms, as well as for Fibonacci-based arrays with perfect periodic auto-correlations, that are zero for all non-zero cyclic shifts. A generalization to include canonical sequences with complex scale factors invokes an equally useful dual form of delta-correlation. We also present $1D$ arrays with a much smaller dynamic range than those where the elements are built using Fibonacci recursion. When Huffman arrays (that are comprised of inherently signed values) are employed as diffuse probe beams for image acquisition, a new two-mask de-correlating step is described here that significantly reduces the total incident radiation dose compared to a prior method that added a positive pedestal-offset.