EDFA noise figure and WSS DGD modeling

The surge in demand for applications such as virtual and augmented reality, expansive cloud services, and high-definition video streaming has significantly increased the need for greater capacity in optical and wireless networks. As these technologies advance, they continually increase their requirements for high bandwidth and low latency, necessitating ongoing improvements in network infrastructure. This is a significant challenge for 5G networks, as the backhaul infrastructure used for previous generations is inadequate for meeting the stringent requirements in terms of very low latency and efficient management of dense traffic. Nowadays, optical systems are approaching their theoretical capacity limits due to significant advancements like probabilistic constellation shaping and forward error correction. These innovations have helped maximize the efficiency of data transmission, pushing towards the Shannon limit. To move forward, the focus must shift to optimizing the physical layer and enhancing the control layer with more efficient resource allocation and management. Innovative approaches such as software-defined networking (SDN), elastic optical networks (EON) and self-driving optical networks have emerged to push performance further. SDN decouples network control from hardware, enabling dynamic management and reconfiguration of resources. EON allows flexible allocation of spectrum based on traffic demands, improving bandwidth efficiency. Self-driving optical networks leverage AI and automation to optimize network operations, improving overall efficiency and reliability. The modeling of Erbium-Doped Fiber Amplifiers (EDFA) is crucial for optimizing optical networks, as these components define the transmission bandwidth and play a crucial role in determining the Optical Signal-to-Noise Ratio (OSNR) of the signal. EDFAs are a well-established technology, yet certain parameters - such as gain, amplified spontaneous emission (ASE) and noise figure - can vary significantly depending on signal attributes such as input power and channel frequency, but these variations do not conform to simple analytical formulas. All of these considerations suggest that the development of accurate predictive models for EDFA parameters could be considered a potentially rewarding challenge. Numerous models for EDFA parameters already exist in the literature, primarily focusing on the estimation of gain. Explicit models have the lowest cost because they can be built based primarily or solely on a-priori knowledge of the devices, but often fall short of high accuracy requirements. Data-driven models, based on neural networks, instead offer greater accuracy but are more difficult to employ in real-world applications due to the large datasets needed for training, which are often unavailable or expensive to acquire. In this thesis, a polynomial model for the noise figure of EDFAs is proposed. The design emphasizes high-speed implementation, to integrate seamlessly with more comprehensive machine learning EDFA models without imposing significant computational delays. Additionally, the model is designed for flexibility, allowing users to adjust the size of the training datasets to balance between accuracy and cost of measurements.