PoT quantization and rotational update mechanism for optimizing equalization in passive optical networks

Abstract

We propose a low-complexity real-time equalization scheme for passive optical networks (PONs), which provides equalization with similar performance and ultra-low digital signal processing (DSP) resource usage compared to traditional schemes in experimental verification. With the help of the proposed scheme, FIR equalizers with multiple taps can be equivalent to equalizers with only two active taps, while other taps are quantified and stored, which can replace complex multiplications with simple shift operations. We introduce powers-of-two (PoT) non-uniform quantization technology to real-time transmission at the physical layer, which helps to save quantization precision and simplify multiplication complexity. We propose the rotational update mechanism (RUM), which solves the problem that the weight value after PoT non-uniform quantization is difficult to update. The proposed scheme with PoT and RUM greatly reduces the required hardware resources and power consumption, and the optical modules can use existing resources to support the additional number of taps required for higher baud rates. We verified the proposed scheme in decision-directed least mean square (DDLMS) equalization. The improved equalization scheme is implemented in the XCVU9P-FLGB2104-2-I Xilinx FPGA with a clock frequency of 230.4 MHz. The results show that the proposed equalization scheme can reduce up to 99.5% DSP resource usage compared to a normal DDLMS equalizer with 15 taps. We successfully realized 29.4912 Gbit/s PAM4 real-time IM/DD transmission over 25-km SSMF with a BER below the KP4 threshold utilizing the proposed DDLMS with PoT and RUM. The innovation of this paper lies in two aspects: (1) the introduction of the PoT quantization method in real-time equalization of PON transmission and (2) proposing RUM for real-time updating of weights in equalizers, which moves the quantification process from offline to real time, making the equalizers more robust for time-varying channels.