Learning method for neural networks using weight perturbation of orthogonal bit sequence and its application to adaptive WDM demultiplexer

• Published in: Journal of lightwave technology Vol. 15; no. 11; pp. 1997 - 2005
• Main Authors: Aisawa, S., Noguchi, K., Miyao, H.
• Format: Journal Article
• Language: English
• Published: New York, NY IEEE 01.11.1997; Institute of Electrical and Electronics Engineers
• Subjects: Applied sciences; Artificial intelligence; Circuits; Computer science; control theory; systems; Connectionism. Neural networks; Degradation; Electric, optical and optoelectronic circuits; Electronics; Exact sciences and technology; High speed optical techniques; Learning systems; Neural networks; Optical computing; Optical fiber networks; Optical transmitters; Signal processing; Wavelength division multiplexing; Learning; Wavelength division multiplexing; Perturbation method; Neural network; Demultiplexer; Performance characteristic
• ISSN: 0733-8724; 1558-2213
• DOI: 10.1109/50.641517

This paper proposes a novel on-chip learning method for hardware-implemented neural networks to achieve an adaptive wavelength division multiplexer (WDM) demultiplexer. The parameters of the neural network are perturbed by orthogonal bit sequences with small amplitude. The parameters are corrected based on the correlation detection result between the perturbed error signal and the corresponding perturbation signal. A learning experiment that transmits 200-Mb/s, four-channel WDM signals through a 40-km fiber and the tracking of the wavelength drift of the optical transmitter successfully demonstrate the proposed method.