Optical Convolution Processing Based on an Amplified Fiber-Optic Recirculating Loop

Convolution processing plays a pivotal role in convolutional neural networks (CNNs), which are widely utilized in image recognition, signal processing, and other applications that demand high computational speed and efficiency. To enhance the computational speed, optical convolution processing (OCP), which leverages the inherent parallelism and high speed of optical systems, has emerged as an effective solution. However, existing OCP implementations often encounter scalability challenges, as most systems are limited to fixed kernel sizes or require substantial hardware expansion to support larger kernels. In this work, we propose an approach to implementing OCP capable of handling various kernel sizes based on an amplified fiber-optic recirculating loop without changing the configuration of the system. For an input data sequence, the multiplication of the input data with a kernel having nn weights is implemented by applying the input data and the kernel weights to a Mach-Zehnder modulator (MZM), to allow the weighted input data to recirculate in the amplified fiber-optic loop nn-11 times. The weighted and time-delayed data are summed at a photodetector (PD), and the convolution operation is completed. The proposed system is evaluated by a proof-of-concept experiment, where convolution operations are performed on the MNIST and fashion-MNIST datasets to generate feature maps using kernels with different sizes. A computational speed of 16 giga operations per second (GOPS) with 92.8% and 73.6% classification accuracies are achieved, respectively.