Key points are not available for this paper at this time.
This paper proposes a truncated parallel architecture for the real-time implementation of noise shaping (NS) for the first time, overcoming the timing constraints imposed by the feedback loop for parallel data processing on an FPGA. Furthermore, a frequency-domain equalization method capable of separating quantization noise while compensating for both channel impairments and residual sampling frequency offset is introduced, significantly improving the accuracy of channel estimation. The performance of the hardware algorithm is experimentally validated through a complete real-time discrete multi-tone modulation (DMT) based intensity modulation with direct detection (IM/DD) system. In contrast to the 3-bit direct quantization system, which exhibits a BER of 7.18 × 10 −3 , the real-time NS system can achieve a significantly lower BER of 1.95 × 10 −4 at a received optical power (ROP) of -13 dBm.
Dong et al. (Mon,) studied this question.