The Price of Sparsity: Sufficient Conditions for Sparse Recovery using Sparse and Sparsified Measurements

We consider the problem of recovering the support of a sparse signal using noisy projections. While extensive work has been done on the dense measurement matrix setting, the sparse setting remains less explored. In this work, we establish sufficient conditions on the sample size for successful sparse recovery using sparse measurement matrices. Bringing together our result with previously known necessary conditions, we discover that, in the regime where $ds\mathrm{/}p\to +\mathrm{\infty }$, sparse recovery in the sparse setting exhibits a phase transition at an information-theoretic threshold of $n_{\text{INF}}^{\text{SP}}=\mathrm{\Theta }(s\log \left(p\mathrm{/}s\right)\mathrm{/}\log \left(ds\mathrm{/}p\right))$, where $p$ denotes the signal dimension, $s$ the number of non-zero components of the signal, and $d$ the expected number of non-zero components per row of measurement. This expression makes the price of sparsity explicit: restricting each measurement to $d$ non-zeros inflates the required sample size by a factor of $\log s\mathrm{/}\log \left(ds\mathrm{/}p\right)$, revealing a precise trade-off between sampling complexity and measurement sparsity. Additionally, we examine the effect of sparsifying an originally dense measurement matrix on sparse signal recovery. We prove in the regime of $s=\alpha p$ and $d=\psi p$ with $\alpha ,\psi \in (0,1)$ and $\psi$ small that a sample of size $n_{\text{INF}}^{\text{Sp-ified}}=\mathrm{\Theta }\left(p\mathrm{/}{\psi }^2\right)$ is sufficient for recovery, subject to a certain uniform integrability conjecture, the proof of which is work in progress.