Constrained Adaptive Sensing

Abstract

Suppose that we wish to estimate a vector x ∈ C n from a small number of noisy linear measurements of the form y = Ax + z, where z represents measurement noise. When the vector x is sparse, meaning that it has only s nonzeros with s ≪ n, one can obtain a significantly more accurate estimate of x by adaptively selecting the rows of A based on the previous measurements provided that the signal-to-noise ratio (SNR) is sufficiently large. In this paper we consider the case where we wish to realize the potential of adaptivity but where the rows of A are subject to physical constraints. In particular, we examine the case where the rows of A are constrained to belong to a finite set of allowable measurement vectors. We demonstrate both the limitations and advantages of adaptive sensing in this constrained setting. We prove that for certain measurement ensembles, the benefits offered by adaptive designs fall far short of the improvements that are possible in the unconstrained adaptive setting. On the other hand, we also provide both theoretical and empirical evidence that in some scenarios adaptivity does still result in substantial improvements even in the constrained setting. To illustrate these potential gains, we propose practical algorithms for constrained adaptive sensing by exploiting connections to the theory of optimal experimental design and show that these algorithms exhibit promising performance in some representative applications.

Ultrafast Ultrasound Imaging as an Inverse Problem: Matrix-Free Sparse Image Reconstruction

“…One solution to this issue may reside in dealing with the coherent operator, by exploring a recent topic on CS, denoted "constrained adaptive sensing", which derives sampling theorems and variants of Theorem 3.1 from [35] where the measurement matrix is more constrained than in standard CS [50].…”

“…(50) Let us now focus on the integral over the variable t which can be formulated as: t∈R n (x t , t) v pe (t − t T x (r) − t Rx (r, x t )) dx t dt = t∈R n (x t , t) u (t T x (r) + t Rx (r, x t ) − t) dx t dt = (n (x t ) * t u) (t T x (r) + t Rx (r, x t )) ,…”

Sparse polynomial chaos expansions via compressed sensing and D-optimal design

“…the authors show that improvements in signal reconstruction accuracy are possible under the assumption that the support (the locations of the non-zero entries) of a signal, with respect to a given basis, is either known or can be estimated. This was the case in their example dealing with Fourier measurements of Wavelet sparse signals [39,Section 4], where it was shown that using an alphabetic optimality criterion from ODE to sequentially choose measurements reduced the meansquared error of signal reconstructions. However, the assumption that prior information regarding the support of c is known in advance is often too restrictive in the context of PC expansions.…”

“…Compared to PC approximations via over-determined least squares approximation (LSA), where the use of ODE has been well explored, ODE for compressed sensing has received less attention. The idea of using ODE for the purpose of compressed sensing, specifically when prior information about the sparsity is known in advance, is not necessarily new [39,40]. For instance, in [39],…”

“…The idea of using ODE for the purpose of compressed sensing, specifically when prior information about the sparsity is known in advance, is not necessarily new [39,40]. For instance, in [39],…”