Low-Complexity Time-Domain DBP Based on Random Step-Size and Partitioned Quantization

Abstract

We propose and experimentally validate a low-complexity time-domain (TD) digital backpropagation (DBP) algorithm for fiber nonlinearity compensation, targeting an optimized hardware implementation. To counteract the coherent accumulation of numerical quantization errors between DBP steps, we propose a random step-size distribution along the optical link (with $\pm$ 5% interval around the optimal step-size). In addition, to further reduce the average quantization bit precision requirements, we propose a partitioned quantization technique, enabling to quantize the finite-impulse response (FIR) filter tail coefficients with significantly lower precision. The proposed low-complexity DBP algorithm is experimentally demonstrated over a 2592 km long-haul wavelength division multiplexing transmission system with 21 $\times$ 32 GBaud PM-16QAM optical channels. Employing the proposed step-size randomization together with dual-time-slot quantization, we demonstrate penalty-free operation at an average of $\sim$ 4 b per FIR coefficient, leading to a 60% complexity reduction when compared to the standard TD-DBP implementation.