Advanced modulation formats for 4 × 100Gbps computing power optical networks and AI-based format recognition

Abstract In the era of the digital economy, enabling users to utilize computing power as conveniently as water and electricity is inseparable from the development of computing power networks. The carrier network of future 6G also needs to achieve deep collaboration between computing and networking. In high-speed computing power optical networks above 400G, nonlinear effects became one of the main factors limiting system transmission performance. Traditional coherent optical communication solutions are complex to implement, require high coherence and stability of light sources, which is difficult, and costly. This paper proposes a low-cost and low-complexity non-coherent solution based on the advanced modulation format Apol-CRZ-FSK to achieve 4 x 100Gbps computing power optical networks. Simulation results show that it exhibits better resistance to non-linear effects compared to traditional modulation formats such as CRZ-FSK and DQPSK, enabling longer single-span and multi-span transmission distances and superior transmission performance. Furthermore, in response to the transmission requirements and signal perception and recognition requirements in future computing optical networks, this study identifies three different modulation formats through the Inception-ResNet-v2 convolutional neural network model. When compared with six deep learning methods including AlexNet, ResNet50, GoogleNet, SqueezeNet, Inception-v4, and Xception, the Inception-ResNet-v2 model achieved the highest accuracy rate of 99.51%, which is a 1.66% improvement over the ResNet50 model. It can provide an effective solution for low-cost, low-complexity, and high-performance signal transmission and signal recognition in the 6G era of high-speed computing power optical networks.