We consider the multi-target detection problem of recovering a set of signals that appear multiple times at unknown locations in a noisy measurement. In the low noise regime, one can estimate the signals by first detecting occurrences, then clustering and averaging them. In the high noise regime, however, neither detection nor clustering can be performed reliably, so that strategies along these lines are destined to fail. Notwithstanding, using autocorrelation analysis, we show that the impossibility to detect and cluster signal occurrences in the presence of high noise does not necessarily preclude signal estimation. Specifically, to estimate the signals, we derive simple relations between the autocorrelations of the observation and those of the signals. These autocorrelations can be estimated accurately at any noise level given a sufficiently long measurement. To recover the signals from the observed autocorrelations, we solve a set of polynomial equations through nonlinear least-squares. We provide analysis regarding well-posedness of the task, and demonstrate numerically the effectiveness of the method in a variety of settings.The main goal of this work is to provide theoretical and numerical support for a recently proposed framework to image 3-D structures of biological macromolecules using cryo-electron microscopy in extreme noise levels.
arXiv:1903.06022v2 [cs.IT] 3 Jun 2019In words: the starting positions of any two occurrences must be separated by at least 2L − 1 positions, so that their end points are necessarily separated by at least L − 1 signal-free (but still noisy) entries in the data. Furthermore, we require that the last signal occurrence in y is also followed by at least L − 1 signal-free entries. This property ensures that correlating y with versions of itself shifted by at most L−1 entries does not involve correlating distinct signal occurrences. Once s is determined, for each position i such that s[i] = 1, one of the signals x k is selected independently at random, and accordingly we set s k [i] = 1. As a result, the only properties of s 1 , . . . , s K that affect the autocorrelations of y (for shifts up to L − 1) are the total number of occurrences of the distinct signals: their individual and relative locations do not intervene. We detail this in Section 3.The Poisson model. If the separation condition is violated, more knowledge about the location distribution is necessary to disentangle the autocorrelations of y. To that effect, we analyze a Poisson generative model. Specifically, for each position i, the number s[i] of signal occurrences starting at that position is drawn independently from a Poisson distribution with parameter γ/L, that is, s[i] i.i.d.