For a compressive sensing model :
$$y_{_{MXN}}=A_{_{MXN}}x_{_{NX1}}$$
where $x$ is $K$ sparse, what is the sufficient condition for Orthogonal matching Pursuit (OMP) to exactly recover the data for both noisy and noiseless case ?
There are multiple statements, like
- $M>Klog(N-K)$ ;
- $M>Klog(N/K)$ ;
- $M>2Klog(N-K)$ ;
- somewhere it's $M>4Klog(N-K)$.
Should there be some constant ratio between $M$ and $N$?