Predictive self-organizing map for vector quantization of migratory signals and its application to mobile...

Area of Science:

• Artificial Intelligence
• Signal Processing
• Communications Engineering

Background:

• Traditional self-organizing maps (SOMs) struggle with time-sequential data exhibiting continuously varying stochastic properties.
• Adaptive vector quantization is crucial for processing dynamic signals in applications like mobile communications.

Purpose of the Study:

• To develop a novel predictive self-organizing map (P-SOM) capable of adaptive vector quantization for migratory time-sequential signals.
• To enhance signal processing by differentiating between stationary noise and time-varying signal components.

Main Methods:

• The P-SOM incorporates weights for both signal values and their time-derivative information to predict future reference vectors.
• Utilizes ordinary weights for stationary noise and time-derivative weights for migrating signal components.
• Applies the P-SOM to a mobile communication receiver employing quasi-coherent detection.

Main Results:

• The P-SOM demonstrates superior adaptive vector quantization by effectively separating signal variations from random noise.
• Simulation experiments show improved bit-error rates (BERs) in a P-SOM adaptive demodulator, particularly in tracking phase rotations due to the Doppler effect.
• Theoretical noise analysis aligns well with experimental findings for both conventional SOM and P-SOM.

Conclusions:

• The P-SOM offers a robust solution for adaptive vector quantization of non-stationary time-sequential signals.
• Its predictive capability, leveraging time-derivative information, significantly enhances performance in dynamic communication environments.
• The P-SOM represents a advancement in signal processing for mobile communication systems facing challenges like Doppler shifts.