Ambiguities in one-dimensional discrete phase retrieval from Fourier magnitudes (Q905612)

The goal of this paper is the characterization of all solutions of the discrete phase retrieval problem, which is generally challenging due to the fact that it is not uniquely solvable. In the one-dimensional continuous setting, the phase retrieval problem can be stated as follows: find a function \(f:\mathbb R\to\mathbb C\) from \(|\mathcal F(f )|\), where \(\mathcal F(f)\) denotes the Fourier transform of \(f\). Section 1 is a survey of the rich literature on the problem of ambiguities in phase retrieval with main emphasis to the one-dimensional case. Sections 2, 3 and 4 contain the original contributions of the authors. Thus, in Section 2, for the first time in literature is obtained a complete mathematical classification of all solutions of the discrete phase retrieval problem for compactly supported complex signals. Moreover, using the zero set of the autocorrelation function, the authors show the main result of this paper, that each nontrivial solution of the discrete phase retrieval problem can be constructed by convolution. In Section 3 one reconsider additional a priori conditions that have been proposed in earlier literature in order to reduce the number of ambiguities or even to ensure uniqueness of the phase retrieval solutions. In particular, the a priori assumption that the desired signal is real and positive, being frequently applied in phase retrieval algorithms, usually does not lead to uniqueness in the one-dimensional case. If besides the Fourier intensities one signal value is known, or alternatively one or more magnitudes of the signal x in the time domain are known, then the authors prove that the signal can be uniquely reconstructed with high probability (up to multiplication with an unimodular constant). Counterexamples to the case when the discrete phase retrieval problem is not uniquely determined by these a priori conditions are constructed. Finally, in Section 4, the authors consider the case when beside \(|\mathcal F(x)(\omega)|^2\), also the intensities of interference signals can be measured. Here one distinguish the cases where the reference signal itself is known or unknown. In the latter case, the authors give a new proof for uniqueness based on their representations of solution ambiguities derived in Section 2 ( see Theorem 4.4). The statements of the main results are too long and complicated in order to be reproduced here.