Source signal sparsity enhancement based on local maximum synchronous extraction transform algorithm for mixed matrix...

Area of Science:

• Signal Processing
• Machine Learning

Background:

• Underdetermined blind source separation (UBSS) faces challenges with suboptimal sparsity and local optima in mixing matrix estimation.
• Traditional time-frequency (TF) methods have limited sparse representation capabilities.
• Clustering algorithms like Fuzzy C-Means (FCM) are sensitive to initial conditions and prone to local optima.

Purpose of the Study:

• To propose a novel mixing matrix estimation algorithm for UBSS systems.
• To enhance source signal sparsity and optimize clustering for improved estimation accuracy.
• To address limitations in existing TF transformation and clustering techniques.

Main Methods:

• Derivation of underdetermined mixing matrix estimation principles based on source sparsity.
• Implementation of a source signal sparsity enhancement algorithm using Local Maximum Synchroextracting Transform (LMSET).
• Adoption of a Proportional-integral-Derivative (PID)-based Search Algorithm (PSA) optimized FCM for robust clustering.

Main Results:

• The proposed LMSET method achieves superior TF resolution and enhanced signal sparsity.
• The PSA-optimized FCM mitigates sensitivity to initial centers and local optima.
• Simulation results show improved TF representation and source signal sparsity across diverse environments.

Conclusions:

• The novel algorithm significantly enhances mixing matrix estimation accuracy in UBSS systems by 19.8%.
• The combined approach of LMSET and PSA-optimized FCM effectively overcomes limitations of traditional methods.
• This work provides a more accurate and robust solution for mixing matrix estimation in underdetermined blind source separation.