Bayesian Compressive Sensing of Sparse Signals with Unknown Clustering Patterns.

Area of Science:

• Signal Processing
• Machine Learning
• Information Theory

Background:

• Compressive Sensing (CS) enables signal recovery from fewer measurements than traditional methods.
• Multiple Measurement Vectors (MMVs) present challenges due to joint sparsity and clustered patterns.
• Existing sparse recovery methods struggle with signals exhibiting unknown clumpiness.

Purpose of the Study:

• To develop an improved sparse Bayesian learning (SBL) method for signal recovery in MMVs.
• To address the challenge of unknown clustered sparsity patterns in MMV signals.
• To enhance the performance of sparse signal recovery by incorporating a novel clumpiness parameter.

Main Methods:

• Proposed a new sparse Bayesian learning (SBL) algorithm.
• Incorporated a total variation-like prior to model the clustering pattern.
• Introduced and learned a new parameter emphasizing clumpiness in the solution supports via a hierarchical SBL algorithm.

Main Results:

• The proposed SBL method effectively recovers sparse signals with unknown clustering patterns.
• The learned clumpiness parameter significantly improves recovery performance in MMV scenarios.
• The algorithm demonstrates effectiveness for both MMV and Single Measurement Vector (SMV) problems.

Conclusions:

• The novel SBL approach offers superior sparse signal recovery for clustered MMVs.
• The introduced clumpiness parameter is crucial for handling unknown sparsity structures.
• This method provides a robust solution applicable to both MMV and SMV signal recovery tasks.