Missing data recovery via a nonparametric iterative adaptive approach

Iteratively reweighted least squares for reconstruction of low-rank matrices with linear structure

“…Initially, MIAA-T declines faster but saturates around ρ = 0.3 and is unable to improve the estimate as more samples are observed. This is consistent with the results presented in [18]. By contrast, IRLS-L keeps reducing the NMSE when more than half of the samples are observed by exploiting the underlying low-rank structure but remains at a certain gap from the oracle CRB.…”

“…Further, we compare with the state of the art iterative adaptive approach for missing temporal data recovery (MIAA-T) [18]. In this approach the data is modeled as…”

“…where a0(ω k ) and a1(ω k ) denote Fourier vectors corresponding to the missing and observed samples, respectively [18]. For the remaining samples, we use the MSE optimal unbiased estimate,θ1 = y ∈ C m .…”

“…For MIAA-T, we follow [18] and set Kg = 10 3 over a uniform grid and terminate after 15 iterations. Fig.…”

SAR imaging via efficient implementations of sparse ML approaches

“…To alleviate this problem, recent work has examined various forms of sparse estimation techniques, such as the sparse learning via iterative minimization (SLIM) method [6], the iterative adaptive approach (IAA) [7], and more recently a set of iterative sparse maximum likelihood-based approaches (SMLA) [8,9]. This class of methods have been found to offer significant performance improvements as compared to the traditional methods not exploiting the sparsity of the signal, generally providing reliable high-resolution estimates with excellent side lobe suppression [10][11][12][13][14]. Yet all these approaches suffer from being computationally cumbersome, which has resulted in a series of recent works focusing on formulating computationally efficient implementations for the SLIM and IAA estimates [15][16][17][18]14].…”