Abstract:We study a blind deconvolution problem on graphs, which arises in the context of localizing a few sources that diffuse over networks. While the observations are bilinear functions of the unknown graph filter coefficients and sparse input signals, a mild requirement on invertibility of the diffusion filter enables an efficient convex relaxation leading to a linear programming formulation that can be tackled with off-the-shelf solvers. Under the Bernoulli-Gaussian model for the inputs, we derive sufficient exact recovery conditions in the noise-free setting. A stable recovery result is then established, ensuring the estimation error remains manageable even when the observations are corrupted by a small amount of noise. Numerical tests with synthetic and real-world network data illustrate the merits of the proposed algorithm, its robustness to noise as well as the benefits of leveraging multiple signals to aid the (blind) localization of sources of diffusion. At a fundamental level, the results presented here broaden the scope of classical blind deconvolution of (spatio-)temporal signals to irregular graph domains.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is to perform blind deconvolution on graphs (networks) in order to locate a small number of source nodes diffusing in the network. Specifically, given a set of graph - signal observations $\{y_i\}_{i = 1}^P$, which are considered as the outputs generated through a diffusion filter (i.e., a polynomial of the known graph - shift operator $S$), the goal is to jointly estimate the filter coefficients $h$ and the input signals $\{x_i\}_{i = 1}^P$ that generate the network observations.
### Problem Description
Since the resulting bilinear inverse problem is ill - posed, the authors assume that the input signals are sparse, which is a reasonable setting because in practical applications, usually only a few seed nodes (source nodes) inject signals and diffuse throughout the network. The location of such source nodes is a challenging problem and has wide applications in multiple fields, including sensor - based environmental monitoring, social networks, neural signal processing, and epidemiology, etc.
### Specific Problems
The paper mainly focuses on the following two aspects:
1. **Blind Deconvolution**: How to recover the filter coefficients $h$ and the input signals $\{x_i\}$ from the observations $\{y_i\}$ without knowing them in advance.
2. **Sparse Source Localization**: How to use graph - signal processing (GSP) techniques to accurately locate sparse source nodes in the network.
### Methods
To solve this problem, the paper proposes the following methods:
- **Convex Relaxation**: By introducing a mild invertibility assumption (i.e., the diffusion filter is invertible), the problem is transformed into a linear programming problem, which can be efficiently solved using off - the - shelf solvers.
- **Stability Analysis**: In the presence of noise, the results of stable recovery are derived to ensure that the estimation error remains controllable.
- **Numerical Experiments**: Through numerical tests on synthetic data and real - world network data, the effectiveness and robustness of the proposed algorithm are verified.
### Main Contributions
- An effective convex relaxation method is proposed, which can achieve blind deconvolution and sparse source localization in the graph domain.
- The exact recovery conditions and stability results are derived to ensure accurate signal recovery in both noiseless and low - noise cases.
- The effectiveness and robustness of the algorithm are verified through numerical experiments, and the advantages in the case of multiple signals are demonstrated.
In summary, this paper aims to solve the blind deconvolution problem on graphs, especially in the application of locating sparse diffusing sources in the network, and provides an effective and stable solution.