In this paper, we implement an optical fiber communication system as an end-to-end deep neural network, including the complete chain of transmitter, channel model, and receiver. This approach enables the optimization of the transceiver in a single end-to-end process. We illustrate the benefits of this method by applying it to intensity modulation/direct detection (IM/DD) systems and show that we can achieve bit error rates below the 6.7% hard-decision forward error correction (HD-FEC) threshold. We model all componentry of the transmitter and receiver, as well as the fiber channel, and apply deep learning to find transmitter and receiver configurations minimizing the symbol error rate. We propose and verify in simulations a training method that yields robust and flexible transceivers that allow-without reconfiguration-reliable transmission over a large range of link dispersions. The results from end-to-end deep learning are successfully verified for the first time in an experiment. In particular, we achieve information rates of 42 Gb/s below the HD-FEC threshold at distances beyond 40 km. We find that our results outperform conventional IM/DD solutions based on two- and four-level pulse amplitude modulation with feedforward equalization at the receiver. Our study is the first step toward end-to-end deep learning based optimization of optical fiber communication systems.
22 February 2019 in Journal of Lightwave Technology
Optics InfoBase is the Optical Society's online library for flagship journals, partnered and copublished journals, and recent proceedings from OSA conferences.
4 December 2017 in Journal of Lightwave Technology
Supporting the ever-increasing data-center-interconnect (DCI) traffic in a cost-effective manner is a great challenge, which requires innovative transmission and digital signal processing (DSP) techniques. Recently, single side-band (SSB) direct-detection (DD) transmissions have been actively considered for data rates beyond 100 Gb/s per channel and distance of hundreds of kilometers due to its capability of electronic chromatic dispersion (EDC) compensation. In addition, several effective DSP techniques to mitigate or suppress the signal-signal beating interference (SSBI) due to the squared-law detection of the photodiode (PD) have been intensively investigated, such as Kramers-Knonig (KK) and SSBI cancellation schemes, showing promising performance at data rates over 200 Gb/s and distance beyond 100 km. In this paper, we demonstrate that high-performance, low-complexity SSB DD transmissions can be achieved by generating a digital carrier (virtual carrier) together with the complex information-bearing signal at the transmitter using only two digital-to-analog converters (DACs). Combining this transmission technique with either the KK field reconstruction or a two-stage SSBI cancellation scheme at the receiver, 8-channel WDM signals with a net data rate of 1.72 Tb/s have been transmitted successfully over a record span length of 200 km at 1550 nm.
We present an experimental study and analysis of a travelling wave series push-pull silicon photonic multi-electrode Mach-Zehnder modulator (ME-MZM) and compare its performance with a single-electrode travelling wave Mach-Zehnder modulator (TWMZM). Utilizing the functionality of the ME-MZM structure plus digital-signal-processing, we report: 1) the C-band transmission of 84 Gb/s OOK modulated data below the KP4 forward error correction threshold with 2 Vpp drive voltage over a distance of 2 km; 2) the transmission of a 128 Gb/s optical 4-level pulse amplitude modulated signal over 1 km of fiber; and 3) the generation of a 168 Gb/s PAM-4 signal using two electrical OOK signals. By comparing the transmission system performance measurements for the ME-MZM with measurements performed using a similar series push-pull TWMZM, we show that the ME-MZM provides a clear advantage in achieving higher baud PAM-4 generation and transmission compared to a TWMZM.
Intensity-modulation with direct-detection (IM/DD) provides a low-cost solution for the data center optical interconnects. To achieve higher data rate, Tomlinson-Harashima precoding (THP) is adopted to increase the operating symbol rate beyond the Nyquist rate defined by the system bandwidth. By applying THP, the signal is precoded by a feedback filter at the transmitter for the mitigation of the inter-symbol interference (ISI) caused by the faster-than-Nyquist (FTN) signaling and the spectral power fading induced by the interaction of the chromatic dispersion (CD) and the direct detection. In this paper, high-performance THP based IM/DD transmission is demonstrated, increasing the achievable data rate using a limited system bandwidth. In the experimental demonstration, we achieve a net bit rate of 185 Gb/s for up to 2 km fiber link, transmitting 74 GBd precoded 8-level pulse amplitude modulation (PAM-8) using an IM/DD system with 33 GHz brick-wall electrical bandwidth limitation.