Digital signal modulation method of photon artificial intelligence computing chip

Abstract

The invention discloses a digital signal modulation method for a photon artificial intelligence calculation chip, and the method comprises the following steps: S01, converting a to-be-recognized optical signal to a preset modal domain, enabling the to-be-recognized optical signal to enter a dispersion compensator, and enabling the dispersion compensator to apply a certain dispersion value to the signal; S02, extracting feature vectors from different modal domains of the to-be-identified signal by using a pre-trained heterogeneous neural network; s03, using a power generation module for mapping the pulse peak power of the input signal under a certain dispersion value to the average power of the output signal; s04, enabling a control and digital signal processing module to drive a dispersion compensator to generate a series of dispersion values; and S05, matching the dispersion-signal average power actual measurement array with dispersion-signal average power samples of various signals with known modulation formats. The method can be directly connected with digital signals of an electronic chip, and quantization errors generated when digital-to-analog conversion is carried out on the digital signals are avoided.

Description

technical field [0001] The invention belongs to the technical field of optical calculation and processing, and in particular relates to a digital signal modulation method of a photon artificial intelligence calculation chip. Background technique [0002] In recent years, thanks to the development of big data, the substantial improvement of computer computing power and the breakthrough of the algorithm itself, deep learning has developed rapidly, and has made great achievements in computer vision, natural language processing, image and video optimization, automatic driving, medical diagnosis and other fields. Delightful results. At the same time, the ever-expanding computing requirements brought about by deep learning have promoted the continuous innovation of processing architectures, and new architectures that can target specific computing have been continuously proposed. [0003] Compared with other architectures, photonic circuits are more suitable for linear operations,...

AI Technical Summary

Problems solved by technology

However, this kind of chip uses analog signals to modulate and calculate light, which has the problems of low precision and poor anti-interference ability, and has high requirements for the dynamic range of electro-optic modulators and photodetectors.

At the same time, AD / DA conversion is required when communicating with digital chips such as FPGAs, which additionally brings quantization errors and increases system costs