Matrix-Form Neural Networks for Complex-Variable Basis Pursuit Problem With Application to Sparse Signal...

Area of Science:

• Computational neuroscience
• Signal processing
• Optimization algorithms

Background:

• The basis pursuit problem is crucial for sparse signal recovery.
• Existing complex-valued neural networks face challenges in stability and convergence.
• Efficient algorithms are needed for complex-variable optimization problems.

Purpose of the Study:

• To propose a continuous-time complex-valued projection neural network (CCPNN) for general complex-variable basis pursuit.
• To develop an improved discrete-time complex projection neural network (IDCPNN) with reduced computational cost.
• To evaluate the performance of the IDCPNN in sparse signal reconstruction using compressed sensing.

Main Methods:

• Development of a novel CCPNN model in matrix state space.
• Theoretical analysis of Lyapunov stability and global convergence for CCPNN.
• Discretization of the CCPNN to create the IDCPNN with a two-step stop strategy.
• Application of IDCPNN to sparse signal reconstruction problems.

Main Results:

• The proposed CCPNN demonstrates Lyapunov stability and global convergence under specific conditions.
• The IDCPNN is theoretically guaranteed for global convergence to the optimal solution.
• The IDCPNN shows superior performance in solution quality and computation time compared to existing methods.

Conclusions:

• The IDCPNN is an effective and efficient algorithm for solving complex-variable basis pursuit problems.
• The IDCPNN offers significant advantages for sparse signal reconstruction in compressed sensing applications.
• This work advances the development of neural network-based optimization techniques.