Polarized signal classification by complex and quaternionic multi-layer perceptrons

Area of Science:

• Signal processing
• Statistical analysis
• Machine learning

Background:

• Polarized signals are crucial in various fields.
• Existing methods may not fully capture the complexities of polarized signals.

Purpose of the Study:

• To propose a statistical framework for polarized signals using quaternionic random processes.
• To evaluate the classification performance of real-, complex-, and quaternionic-valued multi-layer perceptrons (MLPs).
• To investigate the role of class label representations in multi-dimensional neural networks.

Main Methods:

• Development of a statistical framework based on quaternionic random processes.
• Evaluation of multi-layer perceptrons (MLPs) with different data representations (real, complex, quaternionic).
• Analysis of signal-to-noise separation and classification accuracy.

Main Results:

• The proposed quaternionic framework effectively models polarized signals.
• Quaternionic MLPs demonstrate optimal performance in signal-to-noise separation.
• Effective classification of two distinct polarized signals using the proposed methods.

Conclusions:

• Quaternionic random processes provide a powerful tool for analyzing polarized signals.
• Quaternionic MLPs offer superior performance for signal classification and noise separation.
• The study highlights the importance of data representation in neural network applications.