Nearly optimal codebooks from generalized Boolean bent functions over \(\mathbb{Z}_4\) (Q2158234)

Let \(N\geq K>1\) be integers. An \((N, K)\) \textit{codebook} \(C\) is a set \(\{c_0, c_1, \dots , c_{N-1}\}\), where each codeword \(c_i\), \(0 \leq i \leq N -1\), is a unit norm complex vector of length \(K\) over an alphabet, \(A\subset {\mathbb{C}}\). The \textit{maximum cross-correlation amplitude} of \(C\) is defined as \[ I_{\max} (C) = \max_{0\leq i<j\leq N-1} |\langle c_i, c_j\rangle |, \] where \(\langle ... , ...\rangle\) is the usual Hermitian dot product on \({\mathbb{C}}^K\). An old result of Welch states that \(I_{\max}(C)\geq \sqrt{\frac{N-K}{K(N-1)}}\). The goal of this article under review is to construct from generalized bent \({\mathbb{Z}}/4{\mathbb{Z}}\)-valued quadratic forms codebooks \(C\) with \(I_{\max}(C)\) ``nearly achieving'' the Welch lower bound. Sections 1 and 2 of this paper contain background and notation, some of which is recalled below. Fix \(n>1\) and let \({\mathbb{F}}_n=GF(2^n)\) denote the finite field with \(2^n\) elements. Denote the usual trace function by \(tr:{\mathbb{F}}_n \to {\mathbb{F}}_1\) by \(tr\). The \textit{Walsh transform} of a function \(f:{\mathbb{F}}_n\to {\mathbb{F}}_1\) is defined by \(W_f(v)=\sum_{u\in {\mathbb{F}}_n} (-1)^{f(v)+tr(uv)}\), for \(v\in {\mathbb{F}}_n\). The function \(f\) is \textit{bent} if \(|W_f(v)|=2^{n/2}\) for all \(v\in {\mathbb{F}}_n\). Let \({\mathbb{Z}}/4{\mathbb{Z}}\) denote the ring of integers mod \({4}\), let \(h\in {\mathbb{Z}}/4{\mathbb{Z}}[z]\) be a monic irreducible polynomial of degree \(n\), and let \(R= {\mathbb{Z}}/4{\mathbb{Z}}[z]/(h)\). There is a ring homomorphism \(R\to {\mathbb{F}}_n\) induced by the map, \({\mathbb{Z}}/4{\mathbb{Z}} \to {\mathbb{Z}}/2{\mathbb{Z}}={\mathbb{F}}_1\), \(x\mapsto \overline{x}=x\pmod{2}\). The group of units \(R^*\) contains a unique cyclic group of order \(2^n-1\), denoted \(T^*=\langle \xi \rangle\), for some generator \(\xi \in R\). We have \(T^*\cong {\mathbb{F}}_n^*\). Let \(T=T^*\cup \{0\}\) denote the Teichmüller representatives of \(R\). Each \(x\in R\) has a ``\(2\)-adic expansion'', \(x=u+2v\), for some unique \(u,v\in T\). The \textit{trace function} \(Tr:R\to {\mathbb{Z}}/4{\mathbb{Z}} \) is defined by \[ Tr(u+2v) = \sum_{i=0}^{n-1} (u^{2^i}+2v^{2^i}), \] for all \(u,v \in T\). The \textit{Walsh transform} of a function \(f:T\to {\mathbb{Z}}/4{\mathbb{Z}}\) is defined by \(W_f(v)=\sum_{u\in T} (\sqrt{-1})^{f(v)+Tr(uv)}\), for \(v\in T\). The function \(f\) is \textit{generalized bent} if \(|W_f(v)|=2^{n/2}\) for all \(v\in T\). In this notation, section 3 of the paper under review begins the task of determining several explicit families \({\mathcal{F}}\) of quadratic generalized bent functions \(Q:T\to {\mathbb{Z}}/4{\mathbb{Z}}\) and a corresponding codebook \(C=C_{\mathcal{F}}\). Furthermore, each of these codebooks \(C\) has the property that \(I_{\max}(C)\) can be computed precisely and ``nearly achieves'' the Welch lower bound. The constructions of the authors are too technical to be given here but they are carefully explained throughout this section. For details, the interested reader is referred to the paper itself.