Experimental Investigation of Nonlinear Vector Autoregressor-Based Channel Equalization for IM/DD Transmission With Silicon Microring Modulators

To meet the bandwidth and power demands of data center interconnects, multiple microring modulators (MRMs) integrated on a single chip can be combined with efficient digital signal processing to provide an attractive solution. In this work, we experimentally investigate the nonlinear vector autoregressor (NVAR) equalizer applied to a 40 and 50 GBaud PAM-4 MRM-based communication setup. The communication scheme is designed with an electro-optical frequency comb source integrated with a single silicon photonics circuit with 20 cascaded MRMs, of which 4 are transmitted simultaneously in back-to-back (B2B) and 2-km direct detection scenarios. Our results demonstrate that the 2nd-order polynomial degree-based NVAR equalizer outperforms both the reservoir computing (RC) scheme, linear feedforward equalizer, and nonlinear Volterra equalizer across all investigated scenarios, including wavelength-division multiplexing (WDM) with 4 transmitted channels. The NVAR equalizer requires fewer multiplications per equalized symbol compared to the other equalizers. To achieve the best transmission performance the proposed NVAR architecture requires optimization of only 4 critical hyper-parameters resulting in lower optimization complexity compared to RC, which requires the optimization of at least 9 key hyperparameters. With its simple design based on input vector manipulations, low complexity, and relaxed optimization requirements, the NVAR shows significant potential as a high-performance candidate for intra-data center short-reach transmission links.