Photonic analog-to-digital converter powered by a generalized and robust convolutional recurrent autoencoder

Area of Science:

• Photonics
• Signal Processing
• Machine Learning

Background:

• Photonic analog-to-digital converters (PADCs) face challenges with time mismatches, degrading signal fidelity.
• Existing compensation methods may lack robustness to untrained signals or varying system states.

Purpose of the Study:

• To propose and evaluate a convolutional recurrent autoencoder (CRAE) for compensating time mismatches in PADCs.
• To demonstrate the CRAE's generalization capabilities to untrained signals and system states.

Main Methods:

• A convolutional recurrent autoencoder (CRAE) architecture was developed.
• The CRAE was trained using mismatched linear frequency modulated (LFM) signals.
• Performance was evaluated with both LFM and Costas frequency modulated signals under different system states.

Main Results:

• The CRAE effectively compensated for time mismatches ranging from 35 ps to 137 ps.
• The system demonstrated robustness to untrained mismatches and signal types.
• Spur-free dynamic range (SFDR) improved significantly, exceeding 31.6 dBc even with substantial PADC degradation.

Conclusions:

• The proposed CRAE offers a robust and generalized solution for time-mismatch compensation in PADCs.
• This approach enhances PADC performance and reliability across diverse operating conditions.
• The CRAE's ability to handle untrained conditions marks a significant advancement in PADC signal processing.