Real-Time Implementation of Coherent Receiver DSP Adopting Stream Split Assignment on GPU for Flexible Optical Access Systems

We aim to maximize the flexibility of access networks by combining two trends: digital signal processing (DSP)-based physical (PHY) layer technologies and network function virtualization (NFV). We implement an advanced DSP-PHY on a general-purpose server. However, given that the server must handle high data transfer rate, the transfer time is a significant issue for softwarizing DSP-PHY. This article proposes a pipeline architecture that adopts split assignment of the transfer stream; it can execute data transfer and DSP-PHY calculation in parallel which significantly reduces the processing time. We implement on a graphic processing unit (GPU) a 5-Gbit/s coherent receiver DSP handling the output of a 40-Gbit/s analog-to-digital converter. As evaluated in an optical system, our proposal significantly relaxes the processing time requirement for DSP algorithm calculations; the processing time margin is 11.2 times greater than that of previous work. An additional digital filter is also successfully implemented in real time; it offers an optical power budget of 47.2-dB, a gain of 2-dB.