A Photonic Convolution Processor Based on Matched Pair of Arrayed Waveguide Gratings

We demonstrate a photonic convolution processor that uses the spectral shifting in output waveguides for adjacent input waveguides of an arrayed waveguide grating to implement the convolution sliding operation and generate convolution output maps without repetitive matrix calculations. We implement a balanced pair of arrayed waveguide gratings to realize for the first time both negative and positive weighting of photonic convolution kernels. We demonstrate the convolution processor experimentally, integrating a pair of arrayed waveguide gratings with a broadband Mach-Zehnder interferometer on the silicon photonics platform. Experimental results demonstrate the effectiveness of the proposed convolution processor in edge detection.