|
Intensity Modulation and Direct Detection (IMDD) technology, with its low latency, low power consumption, and high throughput advantages, has emerged as the preferred solution for high-efficiency data transmission and exchange between large-scale computing clusters. Digital Signal Processing (DSP) technology can further enhance IMDD transmission performance by compensating for transmission impairments. In this paper, we propose a Functional Link-based Volterra equalizer (FL-Vol) that combines the Functional Link Artificial Neural Network (FLANN) and the Volterra equalizer (Vol). The FL-Vol demonstrates superior nonlinear characterization capabilities, benefiting from higher-order terms in the extended basis functions and cross terms in the Vol. We select the classical Least Mean Square (LMS) criterion for self-adaptive parameter iteration and utilize a variable step-size algorithm based on a modified sigmoid function to balance the rate of convergence and steadystate error. We validate the equalization performance of FL-Vol through a single lane 112Gbps O-band PAM-4 IMDD transmission experiment. Our performance comparison of these equalizers reveals that the 2nd-FL-Vol offers up to a 2dBm improvement in equalization performance with similar computational complexity compared to the 2nd-Vol. Furthermore, the 2nd-FL-Vol reduces computational complexity by approximately 30% while maintaining similar equalization performance compared to the 3rd-Vol. Notably, despite the theoretical advantage of the 3rd-Vol structure in nonlinear characterization, the 2nd-FL-Vol consistently outperforms the 3rd-Vol, especially at higher received optical powers. Additionally, as expected, the variable step-size algorithm improves the performance of all equalizers.
|
