Practical Kramers-Kronig Phase Retrieval FIR Filter With the Gibbs Phenomenon

Kramers-Kronig (KK) receiver shows strong feasibility since the phase can be extracted from the detected intensity through the Hilbert transform (HT). To reduce the complexity and make the HT more suitable for the practical KK system, the HT can be approximated as the finite impulse response (FIR) filter, i.e., the HT-FIR filter. However, the finite Fourier series expansion of the HT-FIR filter suffers from oscillations around the discontinuities and further results in an unsatisfactory in-band flatness, which is the so-called Gibbs phenomenon. In the experiment and the simulation, we found that the performance of the KK system using the HT-FIR filter is strongly impaired due to the Gibbs phenomenon, by the roll-off factor of the shaping filter, frequency gap between the direct-current (DC) component and one edge of the signal spectrum, carrier-to-signal power ratio (CSPR), tap number of the HT-FIR filter, and the digital upsampling rate. We have systematically investigated the practical KK phase retrieval performance and the computational complexity considering the Gibbs phenomenon due to the adoption of the HT-FIR filter, rather than the HT used in the previous works without the Gibbs phenomenon. We conclude that there is an optimal upsampling rate when the HT-FIR filter is used, and the optimal upsampling rate is closely related to the five aspects mentioned above. We have carried out a transmission experiment using a 112-Gbit/s 16-QAM system over 1920-km with a Raman fiber amplifier to investigate the limit transmission performance of the practical KK receiver. The experimental results show that by using the 9-tap HT-FIR filter with a digital upsampling rate of 84-GHz (3-
sps
), the signals can be transmitted the same distance as the HT, up to 1440-km with a bit error ratio (BER) below the 20% soft-decision forward error correction threshold.