Multi-task photonic time-delay reservoir computing based on polarization modulation.

Area of Science:

• Optoelectronics
• Reservoir Computing
• Nonlinear Dynamics

Background:

• Reservoir computing (RC) offers a powerful framework for complex system modeling and prediction.
• Photonic implementations of RC promise high-speed information processing capabilities.
• Multi-task learning in RC requires adaptable nonlinear activation functions.

Purpose of the Study:

• To propose and experimentally demonstrate a multi-task photonic time-delay reservoir computing (RC) system.
• To utilize polarization modulation for electrical-to-optical conversion and nonlinear operation.
• To investigate the impact of adjustable nonlinear bias on task-specific performance and information processing capacity (IPC).

Main Methods:

• A novel RC system employing a polarization modulator (PolM), polarization controller (PC), and polarizer as an equivalent Mach-Zehnder modulator (MZM).
• Electrical-to-optical conversion and nonlinear function generation achieved through the equivalent MZM.
• Optimization of the equivalent MZM bias for distinct tasks to enhance multi-task performance and evaluate task-independent IPC.

Main Results:

• Experimental validation of the photonic time-delay RC system for multi-task processing.
• Demonstration that readout bias adjustment significantly impacts IPC and task performance.
• Achieved high accuracy on benchmark tasks: NARMA10 (NMSE=0.2103) and IPIX radar signal prediction (NMSE=0.0031) at 1.06 Mb/s.

Conclusions:

• The proposed polarization-modulated photonic time-delay RC system is effective for multi-task learning.
• Adjustable nonlinear bias is crucial for optimizing performance across different tasks.
• This approach offers a promising avenue for high-speed, adaptable photonic computing.