Construction of robust frames in erasure recovery (Q2348017)

The paper is concerned with the problem of recovering signals from frame coefficients, specifically in situations when some coefficients are missing at unknown locations (being, for example, lost in transmission) or the coefficients are not in the original order. The authors present some explicit construction of almost robust and almost self-located frames, introduced in [\textit{D. Han} and \textit{W. Sun}, IEEE Trans. Inform. Theory 60, 4013--4025 (2014)]. Such systems allow the recovery of erased frame coefficients at unknown locations and the correct order of coefficients from any of their rearrangements for almost all signals. More precisely, a frame \(\{\varphi_{i}\}_{1\leqslant i\leqslant N}\) for a finite dimensional Hilbert space \(\mathcal H\) is called almost robust with respect to \(m\)-erasures if any \(f\in \mathcal H\setminus \mathcal H_{0}\) can be recovered from its frame coefficients with \(m\)-erasures at unknown locations, where \(\mathcal H_{0}\) is the union of finitely many proper subspaces of \(\mathcal H\) and therefore is of measure zero, whilst the frame is called almost self-located if the sequence of frame coefficients \(\{\langle f,\varphi_{i}\rangle\}_{1\leqslant i\leqslant N}\) can be recovered from any of its rearrangements for any \(f\in \mathcal H\setminus \mathcal H_{0}\). Two main results of the paper are the following. Let \(p_{1},\dots, p_{N}\) be \(N\) different prime numbers, where \(N-2\geqslant n\geqslant 2\). Set \(\varphi_{i}=(1,\sqrt{p_{i}},\dots, (\sqrt{p_{i}})^{n-1})^{T}\), \(1\leqslant i\leqslant N\). The first result states that \(\{\varphi_{i}\}_{1\leqslant i\leqslant N}\) is a frame for \(\mathbb R^{n}\) which is both almost self-located and almost robust with respect to \((N-n-1)\)-erasures. Let now \(M\) be an \((N-n)\times N\) matrix with \((i,j)\) entry being \((\sqrt{p_{j}})^{i-1}\). Suppose that \(g_{1},\dots, g_{n}\) form a basis for \(\mathcal N(M)=\{x: Mx=0\}\), where \(g_{i}=(a_{i,1},a_{i,2},\dots, a_{i,N})^{T}\), \(1\leqslant i\leqslant n\), and let \(\psi_{j}=(a_{1,j},a_{2,j},\dots, a_{n,j})^{T}\), \(1\leqslant j\leqslant N\). The second result states that\(\{\psi_{i}\}_{1\leqslant i\leqslant N}\) is a frame for \(\mathbb R^{n}\) which is both almost self-located and almost robust with respect to \((N-n-1)\)-erasures.