Adaptive Partial-Response Neural Network Equalization for Bandwidth-Limited PAM Transmission in Intra-Datacenter Interconnect

High-speed optical transmission systems suffer from severe intersymbol-interference (ISI) caused by limited bandwidth. To address this challenge, previous studies on bandwidth limited signals have explored the use of a post filter after full-response equalization to whiten colored noise and enhance system performance. However, the post filter requires additional training or scanning processes to the appropriate coefficient values, which introduces additional overhead and time consumption. A low-complexity duobinary neural network equalization (DB-NNE) has been investigated recently, which is trained with the target of the duobinary version of the original symbol. Although this approach achieves both channel equalization and noise suppression, it may not achieve the optimal coefficient value for noise suppression. In this research, we propose an adaptive partial-response neural network equalization (APR-NNE) for bandwidth-limited signal processing and nonlinearity mitigation. Our approach seamlessly integrates channel equalization and noise suppression, achieving an optimal balance between these two functions. To verify the effectiveness of the proposed scheme, we experimentally demonstrate the transmission of beyond-150-GBaud PAM-4 signals in a system with −10-dB bandwidth of 57 GHz. The experimental results show that the APR-NNE can compensate the bandwidth-limitation induced ISI effectively, enabling the achievement of the 7% HD-FEC threshold for 155-GBaud PAM-4 signals over a 0.5-km SSMF in C-band.