研究目的
To calibrate 2×2 dual-ring assisted Mach-Zehnder interferometer switches based on a back-propagation artificial neural network (BP-ANN) for effective calibration of DR-MZI switch elements in a large-scale switch fabric.
研究成果
The BP-ANN based algorithm effectively calibrates the 2×2 DR-MZI switch, enabling configuration to cross and bar states at four operation wavelengths. This method presents a promising solution for automatic control of large-scale silicon optical switch fabrics.
研究不足
The accuracy of phase identification is limited (~ 0.01π), affecting the extinction ratio at the bar-state. The wavelength error is about 100 pm, influenced by thermal tuning efficiency and resonant wavelength drift in an ambient environment.
1：Experimental Design and Method Selection:
The study employs a BP-ANN to identify the differential phase of the two rings in a DR-MZI switch from transmission spectra.
2：Sample Selection and Data Sources:
Spectrum samples are generated from the transfer matrix model, with parameters varied to account for fabrication non-uniformity.
3：List of Experimental Equipment and Materials:
Includes an optical dispersion and loss analyzer from Agilent for measuring transmission spectra.
4：Experimental Procedures and Operational Workflow:
Involves pre-tuning the element to the target wavelength, using BP-ANN for phase identification, and applying calculated voltages to microheaters for desired spectrum.
5：Data Analysis Methods:
The BP-ANN is trained using the Adam optimizer and mini-batch gradient descent method, with performance evaluated based on the mean square error between output parameters and real values.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
