Frequency-Dependent Variations in DAC ENoB: Predicting Accurate High-Speed Eye-Diagrams

Abstract

While previous digital-to-analog converter (DAC) simulation models predict frequency-dependent effective number of bits (ENoB) well, the time domain distortions are exaggerated. We introduce a model preserving ENoB accuracy and it improves the treatment of time-domain behavior. We enhance model accuracy by filtering out spurious frequency content from aliasing and distortion that mask true behavior. We validate frequency-dependent ENoB for three commercial DACs by comparing our model output to known performance. We run a back-to-back electrical experiment to assess the accuracy of the model to predict bit error rate (BER) for baud rates from 70 to 120 Gbaud. We show that, unlike our model, the previous model leads to predictions of eye-opening and BER that are overly pessimistic at higher baud rates. We analyze the influence of oversampling, showcasing the trade-offs between the accuracy of the expected ENoB and the computational complexity. This work provides a comprehensive tool for understanding the constraints imposed by DACs, facilitating the optimization of next-generation optical communication systems.