Advanced Digital Signal Processing for 400 Gb/s/$\lambda$ IM/DD Intra Data Center Systems

This work presents an experimental study of 400 Gb/s/$\lambda$targeted IM/DD systems in the O-band, which are of high relevance for next-generation 3.2 Tb/s short-reach Ethernet connections inside data centers. For such systems, advanced digital signal processing (DSP) has become essential to overcome bandwidth limitations and accumulated dispersion as dominant channel impairments. On a 2 km link, PAM-4, -6, and -8 as modulation formats, the use of pre-emphasis in the power-constrained transmitter, as well as precoding with the objective to break error bursts are analyzed. At the receiver, four DSP schemes are studied, including (1) full-response Volterra nonlinear equalization, (2) partial-response equalization implemented as Volterra equalization combined with a post-filter for channel shortening and subsequent sequence estimation, (3) duobinary-targeted equalization, combined with sequence estimation for duobinary decoding and (4) a machine-learning-based equalization and sequence estimation architecture. The experimental study is extended to 10 km transmission at wavelengths between 1290-1330 nm to quantify the impact of chromatic dispersion and define permissible wavelength ranges. Optimal DSP architectures are determined, depending on the modulation format and the signaling rate. The results demonstrate achievable information rates of 437 Gb/s for 160 GBd PAM-8, 412 Gb/s for 174 GBd PAM-6, and 402 Gb/s for 210 GBd PAM-4 signaling.