This paper proposes a nonlinear equalization technique enabled by long short-term memory (LSTM) recurrent neural networks. The proposed technique is implemented at the end of offline digital signal processing. And two approaches utilizing the LSTM network are experimentally tested and demonstrated in transmission of a 50-Gb/s four-level pulse amplitude modulation intensity modulation direct detection link over 100-km standard single-mode fiber. The first approach uses the LSTM network-based equalizer to directly categorize the received signal into four amplitude levels, and the second approach uses the LSTM network to estimate signal noise for compensating the received signal. The experimental results show remarkable performance improvement of the proposed method over conventional linear equalizers, and significant enhancement at high launch power compared with Volterra filtering. Also, the proposed method reveals better short-time universality.

High-performance germanium photodiodes are crucial components in silicon photonic integrated circuits for large-capacity data communication. However, the bandwidths of most germanium photodiodes are limited by the intractable resistance–capacitance parasitic effect. Here, we introduce a unique U-shaped electrode to alleviate this issue, reducing the parasitic effect by 36% without compromising any other performance. Experimentally, a large bandwidth of 103 GHz, an optical responsivity of 0.95 A/W at 1550 nm, and a dark current as low as 1.3 nA are achieved, leading to a record high specific detectivity. This is the first breakthrough to 100 GHz bandwidth among all vertical germanium photodiodes, to the best of our knowledge. Open eye diagrams of 120 Gb/s on-off keying and 200 Gb/s four-level pulse amplitude signals are well received. This work provides a promising solution for chip-based ultra-fast photodetection.

In this letter, a bidirectional symmetric 100 Gb/s/ $\lambda $ coherent passive optical network (PON) scheme with a simplified optical network unit (ONU) transceiver is proposed and experimentally demonstrated. In the proposed scheme, the downstream signal is generated in a dual-polarization coherent transmitter with Alamouti-coding at the optical line terminal (OLT) side and then received by a single-polarization heterodyne receiver at the ONU side. The upstream signal is generated in a low-cost C-band electro-absorption modulated laser (EML) at the ONU side and then received by a standard dual-polarization integrated coherent receiver at the OLT side. The experimental results show that more than 30 dB power budget is achieved after 25 km standard single mode fibre transmission for both downstream and upstream signal at 100 Gb/s/ $\lambda $ . The cost at the ONU side is maintained low as the required bandwidth of the optoelectrical device is >25 GHz and the local oscillator power at the ONU side can be as low as 4 dBm.

In this paper, a novel low-complexity adaptive channel equalization (ACE) algorithm for digital coherent optical systems is proposed and experimentally demonstrated. We divide the conventional N-tap butterfly ACE into two N-tap polarization independent filters and a 1-tap butterfly adaptive equalization filter. The computational complexity can reduce about 40% of multiplier operations in the digital signal processing (DSP). We evaluate the effectiveness of our proposed ACE algorithm in a 10-Gb/s real-time coherent transmission platform. It is shown that our proposed ACE algorithm has similar performance as conventional ACE algorithm and better polarization tracking ability.

A novel method assisted by feed-forward artificial neural network model for joint identification of in-phase/quadrature (IQ) time-skews and power-imbalances for coherent optical transmitters is proposed in this paper. This method not only reduces the complexity of hardware design but also significantly improves the accuracy of time-skew/power-imbalance estimation, therefore largely enhances the efficiency of coherent transmitter offline calibration. The proposed method works in the heterodyne detection manner to detect both the time- and power- imperfections. A modified 2 × 2 real-valued channel equalizer for QPSK and 16-QAM signals is applied to extract the channel coefficients, which can fully reconstruct the signals. A key tool using artificial neural networks is used to establish the explicit numerical relationships between the values of IQ time-skew/power-imbalance and the channel coefficients. Both simulation and experimental tests are carried out to verify the capability of the proposed method. Simulation results show that the mean square errors (MSE) of IQ time-skew and power-imbalance estimation can reach below 0.03% of the symbol period and 5.72 × 10⁻⁵ at the optical signal-to-noise ratio value of 20 dB. Experiment tests based on 100-Gb/s and 200-Gb/s coherent optical modules show that the mean absolute errors (MAEs) of estimated IQ time-skew and power-imbalance are 0.145 ps and 0.01 for 32-GBaud QPSK signal and 0.162ps and 0.006 for 32-GBaud 16-QAM signal. A demonstrative calibration process is applied to a coherent optical module with pre-set IQ time-skew and power-imbalance by improving the Q² factor of 32-Gbaud 16-QAM signals from 12.9 dB to 20.3 dB in a coherent optical transmission link at back-to-back case.