On-Line Learning of Write Strategy for Ultra-Speed CD-RW Optical Recorder.

Area of Science:

• Optical Data Storage
• Machine Learning Applications
• Information Technology

Background:

• CD-RW recorders face challenges in maintaining consistent recording performance across discs with varying recording histories (1st, 2nd, or 3rd use).
• A single set of write strategy parameters is insufficient to meet recording specifications for these diverse disc conditions.

Purpose of the Study:

• To develop an effective on-line machine learning approach for optimizing the write strategy of infrared diodes in ultra-speed CD-RW recorders.
• To address the convergence stagnation issue encountered by conventional genetic algorithms (GAs) in this application.

Main Methods:

• Integration of a genetic algorithm (GA) with jitter measurements for on-line learning of write strategies.
• Proposal and implementation of a novel dynamic parameter encoding scheme to enhance GA performance.
• Evaluation of the proposed method's ability to satisfy recording specifications for discs with different recording histories.

Main Results:

• The proposed dynamic parameter encoding scheme significantly improves GA convergence compared to conventional methods.
• The enhanced GA demonstrates superior exploration of the search space for optimal write strategy parameters.
• The approach effectively addresses the challenge of satisfying recording specifications across multiple disc recording cycles.

Conclusions:

• The dynamic parameter encoding scheme is a viable solution for optimizing CD-RW write strategies in ultra-speed recorders.
• This machine learning approach offers improved performance and adaptability for optical data storage technologies.
• Further research can explore this method for other adaptive recording systems.