Stable Recovery of Sparse Signals via $l_p-$Minimization

Abstract

In this paper, we show that, under the assumption that $\|\e\|_2\leq ε$, every $k-$sparse signal $\x\in \mathbb{R}^n$ can be stably ($\epsilon \ne 0$) or exactly recovered ($\epsilon =0$) from $\y=\A\x+\e$ via $l_p-$mnimization with $p\in (0,\bar{p}]$, where \beqnn \bar{p}= \begin{cases} \frac{50}{31}(1-δ_{2k}), &δ_{2k}\in[\frac{\sqrt{2}}{2}, 0.7183)\cr 0.4541, &δ_{2k}\in[0.7183,0.7729)\cr 2(1-δ_{2k}), &δ_{2k}\in[0.7729,1) \end{cases}, \eeqnn even if the restricted isometry constant of $\A$ satisfies ${\delta }_{2k}\in [\frac{\sqrt{2}}{2},1)$. Furthermore, under the assumption that $n\le 4k$, we show that the range of $p$ can be further improved to $p\in (0,\frac{3+2\sqrt{2}}{2}(1-{\delta }_{2k})]$. This not only extends some discussions of only the noiseless recovery (Lai et al. and Wu et al.) to the noise recovery, but also greatly improves the best existing results where $p\in (0,\min \{1,1.0873(1-{\delta }_{2k})\})$ (Wu et al.).

Cited by in corpus (1)

• A unified recovery bound estimation for noise-aware Lq optimization model in compressed sensing