1. Introduction

The migration of bandwidth-intensive services and applications to the cloud continues to drive the development of high speed Data Center Interconnects. 200 and 400 gigabit Ethernet (GbE), which uses CWDM to provide 50 Gb/s on up to 8 optical carriers, is encapsulated in the IEEE 803.3bs standard [1] for up to 10 km transmission. In recognition of the critical requirements for cost-efficiency and scalability associated with DCI transceiver design, 200/400 GbE calls for the use of Intensity Modulation and Direct Detection (IM/DD) with 4-level Pulse Amplitude Modulation (PAM-4); providing high capacity and relatively simplistic optical links. As required DCI speeds evolve beyond 400 Gb/s to 800 Gb/s, and even >1 Tb/s, over the coming years, the onus on a cost-effective and scalable WDM platform must be maintained [2]. This places a great deal of importance on the development of small-footprint and low-cost multi-wavelength sources capable of providing high channel count, and spectrally efficient, DCIs.

To overcome the effects of Chromatic Dispersion (CD), current commercial short-reach/DCI deployments make use of O-band transmission. Up to 8 wavelengths, spaced by 800 GHz, are used to carry 50 Gb/s each - giving a spectral efficiency of < 0.1 b/s/Hz. Clearly, the introduction of a Dense WDM (DWDM) grid and a higher channel count would help to greatly increase spectral efficiency and facilitate future DCI capacity scaling. However, this is hindered in the O-band because the low dispersion in this transmission window increases the relative impact of Four Wave Mixing (FWM) [3]. As such non-linearity can be highly detrimental to the performance of multi-wavelength systems employing a narrow channel spacing, increasing the spectral efficiency of DCIs through DWDM networking can more easily facilitated by transmission in the C-band [4].

WDM interconnects are commonly realised using an array of on-chip single-carrier optical sources such as Distributed Feedback (DFB) [5] lasers or Vertical Cavity Surface Emitting Lasers (VCSEL) [6]. Such a transceiver implementation limits scalability as higher channel counts are required for terabit-scale transmission. These issues can be alleviated by the deployment of multi-wavelength optical sources [2]. Optical gain switching [7] provides multiple wavelengths from a single DFB, but requires a relatively high powered RF source which limits the scale of power and footprint reduction. Another approach is to make use of the multi-wavelength output of a Mode Locked Laser (MLL) whose compact and relatively simplistic implementation can provide a very high channel count with a fixed carrier spacing [8]. In Ref. [8], we have previously outlined an envisaged chip-scale transceiver design which, based on a hybrid photonic integration approach, contains a Q-Dash MLL and a Silicon Photonic (SiP) micro-ring structure for (de)multiplexing of wavelength channels. By harnessing the power efficiency provided by the MLL and SiP [9] components, this approach has the potential to provide a relatively low-powered transceiver solution.

Other previous works have shown the suitability of MLLs for DCI transmission, and their compatibility with surrounding photonic transceiver technologies [10–12]. Ref. [10] demonstrates Nyquist-filtered 32 GBaud transmission on 64 optical carriers derived from an O-band quantum-dot MLL, with each tested carrier being filtered, modulated and transmitted independently. Also using a quantum-dot laser, Ref. [11] shows real-time WDM transmission of 28 GBaud PAM-4 on 4 carriers, simultaneously, on a 50 GHz grid in the C-band, over 80 km of fiber. Recently, Ref. [13] has reported the use of a 34.22 GHz Q-Dash MLL to perform 28 GBd PAM-4 C-band transmission on 48 carriers, again, on an individual basis. Our previous work has demonstrated the use of a Q-Dash MLL for DCI transmission using both Orthogonal Frequency Division Multiplexing (OFDM) [14] and PAM-4 [12]. Ref. [12] shows how digital signal encoding of PAM signals can be employed to overcome the significant low frequency Relative Intensity Noise (RIN) associated with Q-Dash MLLs - in this instance facilitating the transmission of 28 GBaud PAM-4 transmission on 39 modes, individually.

In this work, a single section, three layer Q-Dash MLL with a channel spacing of 32.5 GHz is shown to support Nyquist-filtered 28 GBaud PAM-4 transmission on modes spanning >1 THz frequency range, and without the requirement for additional digital encoding of the PAM sequence. Additionally, in order to truly evaluate the device’s suitability within in the context of DWDM C-band transmission, this work builds on the prior demonstrations by experimentally undertaking 400 Gb/s (8$\times$56 Gb/s) transmission; i.e. the 8 modulated carriers spaced by 32.5 GHz are transmitted simultaneously in a constructed DWDM experimental testbed. With Bit Error Rates (BER) below the Hard-Decision Forward Error Correction (HD-FEC) limit exhibited for all of the 8 adjacent wavelength channels, the successful 400G DWDM transmission gives a spectral efficiency of 1.54 b/s/Hz.

2. Q-Dash mode-locked laser

The device used was a single section MLL whose gain medium consisted of three layers of InAs Q-Dash. Generic details of the material growth by Gas Source Molecular Beam Epitaxy (GSMBE) and device structure can be found in Ref. [15].

A single transverse mode is guided inside the Fabry-Pérot cavity using a buried ridge waveguide structure. The laser mirrors were made using cleaved facets, and the chosen cavity length resulted in a repetition rate of 32.5 GHz. The value of this repetition rate is controlled by the precision of cleavage of the laser mirrors which amounts to about 10 $\mu$m with the current standard processing, resulting in $\sim$100 MHz uncertainty in the FSR of the MLL. While temperature tuning does allow for slight adjustments to this value, a more precise definition of the laser cavity length could potentially be achieved through the development of new lithography techniques, which we are currently investigating. A total (fiber-coupled) output power of $\sim$9 dBm was observed for a bias current of 300 mA. For these conditions, the output optical spectrum (which can be seen in Ref. [16]) exhibited $\sim$45 comb lines within a 3 dB power variation.

Successful multi-level signalling is enabled by this Q-Dash MLL due to its relatively low RIN characteristic - particularly at low frequencies. Characterization of the Q-Dash MLL’s RIN was performed and Fig. 1 shows the RIN profile of an individually filtered mode at 194.942 THz. The figure indicates that the RIN levels exhibited on an individually filtered mode is $\leq$ −135 dBc/Hz at frequencies above 1.6 GHz. Below this frequency, the impact of Mode Partition Noise (MPN) results in an increase in RIN, up to −124 dBc/Hz. Nevertheless, this figure represents a 15 dB reduction in RIN compared to Ref. [12] which required additional signal coding (spectral shaping) to circumvent low frequency noise. This MPN reduction can be explained by the smaller number of Q-Dash gain layers with reduced optical confinement factor - contributing to a reduction in differential gain. This results in a lower value of relaxation oscillation frequency, which in turn contributes to a reduction in MPN enhancement at low frequencies [17].

 

Fig. 1. RIN profile of the filtered wavelength channel at 194.942 THz.

Download Full Size | PDF

It is also worth noting that, compared to two-section InAs/InP MLLs consisting of saturable absorber and gain sections, our previous work has shown that single section devices exhibit improved performance in terms of output power and optical bandwidth [18].

3. Experimental setup

3.1 Optical system

The experimental setup shown in Fig. 2 is designed to evaluate both single channel and WDM transmission scenarios. The multi-wavelength output from the 32.5 GHz Q-Dash MLL, which is biased at 300 mA (resulting in an output power-per-line of about −8 dBm for the 8 comb lines selected for WDM transmission tests), is passed to a Wavelength Selective Switch (WSS) which is configured to select the desired wavelength channel(s) for modulation. The channel(s) are then amplified by an Erbium Doped Fibre Amplifier (EDFA) in order to overcome coupling losses to the unpackaged MLL. A Polarization Controller (PC) sets the polarization to the input of a 40 GHz Mach-Zehnder Modulator (MZM). In the case where 8 channel DWDM transmission is performed, a tunable Delay Interferometer (DI) is used to filter and de-correlate odd and even WDM channels. The modulated signal(s) are transmitted through 1 km of Single Mode Fiber (SMF), and then filtered by a narrow-band tunable Optical Band-Pass Filter (OBPF, Yenista XTM-50) whose 3 dB bandwidth is manually set close to that of the signal (28 GHz) in order to select a single channel for performance evaluation. Additional losses due to the DI used for decorrelation of adjacent channels (in the WDM transmission case) are compensated using a receiver EDFA which boosts the selected wavelength channel power. A wide-band 2nm OBPF is used to supress Amplified Spontaneous Emission (ASE) only, and does not impact WDM channel selection. A Variable Optical Attenuator (VOA) is used to set the power falling on a 30 GHz PIN receiver.

 

Fig. 2. Experimental setup for 28 GBaud PAM-4 transmission using the Q-Dash MLL. The DI is incorporated into the system for WDM transmission. Insets show the transmitted electrical spectrum of the PAM signal at the AWG, and a received optical spectrum of a single modulated WDM channel.

Download Full Size | PDF

3.2 Signal properties

The 28 Gbaud PAM-4 sequences are generated digitally and Nyquist pulse shaped using a Raised Cosine (RC) filter with a roll-off factor of 0.12 - effectively leading to a 500 MHz guard-band between optical channels. Nonlinear pre-distortion is applied to compensate for the nonlinear MZM transfer characteristic. The PAM signals are then up-sampled and loaded into an Arbitrary Waveform Generator (AWG) sampling at 84 GSa/s. The output electrical signal is amplified and then applied to the single-ended MZM. An example transmitted electrical spectrum is shown as an inset in Fig. 2.

At the receiver side, a Real Time Oscilloscope (RTS) sampling at 100 GSa/s is used to capture the PAM-4 signals with down-sampling, timing synchronization and equalization being performed offline. The adaptive equalizer used is a 21-tap Finite Impulse Response (FIR) filter, with tap weights updated using a Decision Directed Least Mean Square (DD-LMS) algorithm.

4. Results and discussion

4.1 Single-channel transmission

The emission spectrum from the Q-Dash laser is shown in Fig. 3(a). In order to demonstrate operation across a wide frequency range, a selection of available modes ranging from 194.584 to 195.856 THz (indicated by a star marker in the figure) are individually tested with PAM-4 transmission at 28 Gbaud. In each case the transmitter WSS and receiver OBPF were tuned to filter a single channel only, and the received optical power was set to 0 dBm. The performance of each channel, in terms of BER, is shown in Fig. 3(b), with the legend showing the precise frequency of each tested channel. With the exception of the channel at 194.584 THz, performances below the HD-FEC limit of 3.8$\times$10$^{-3}$ are achieved, with the best performing channel exhibiting a BER of 1.4$\times$10$^{-4}$. Deviation of BER across the tested channels is attributed to the varying power levels of each Q-Dash output mode relative to the system noise floor, which can be observed in 3(a). These results indicate the Q-Dash MLL’s capability to successfully transmit 50 Gb/s/$\lambda$ over a spectral range of 1.142 THz. Modulation of all 36 wavelength channels in this range would result in a total raw throughput of 2 Tb/s. The inset of 3(b) shows an example eye diagram received on the 194.942 THz channel.

 

Fig. 3. (a) shows the full emission spectrum of the Q-Dash MLL with individually tested channels denoted with a star. The received BER of the corresponding channels is given in (b) with the inset showing the received PAM-4 eye diagram at 194.942 THz.

Download Full Size | PDF

As discussed, the relatively high MPN typically exhibited by Q-Dash MLLs, gives rise to low frequency RIN which can impact the performance of multi-level advanced modulation signals [12]. The results in Fig. 3(b) indicate that the RIN levels exhibited by this Q-Dash MLL were sufficiently low to support PAM-4 modulation without the requirement for additional spectral shaping (line coding) at the transmitter.

4.2 WDM transmission

To evaluate the Q-Dash MLL’s compatibility with C-Band DWDM networking, transmission of a 400 Gb/s (8$\times$28 Gbaud PAM-4) WDM signal was carried out using the full setup shown in Fig. 2. Figure 4(a) shows the optical spectrum at the output of the fiber with 32.5 GHz spaced DWDM channels, from 194.979 to 195.204 THz, labelled 1-8. Figure 4(b) gives the measured BER on each channel, with all values below the HD-FEC limit, and with the inset showing an example received 28 Gbaud PAM-4 eye diagram on channel 4. The best performing channel (8) exhibits a BER of 1.2$\times$10$^{-3}$, while the worst performing channel (5) gives performance at 2.5$\times$10$^{-3}$. The optical spectrum of a received filtered single WDM channel is shown as an inset in Fig. 2.

 

Fig. 4. (a) shows the modulated 8 channel WDM spectrum at the output of the fiber and (b) gives the BER per channel with the inset showing the received PAM-4 eye diagram on channel 4.

Download Full Size | PDF

Considering the modest launch power (total power $\sim$ +5dBm), transmission distance and operation in the C-band, fiber nonlinearities (such a cross phase modulation (XPM) and FWM) do not have any significant impact on DWDM transmission performance. Performance, compared to the single channel transmission case, is degraded mainly due to the reduction in Optical Signal-to-Noise Ratio (OSNR) as a consequence of WDM operation. Specifically, this is caused by the reduction in the optical power per channel as the multi-wavelength signal is now passed to the transmitter EDFA. Further loss is introduced by the DI, while the required receiver EDFA increases optical noise presence in the system. Nevertheless, successful 400 Gb/s transmission is achieved, giving a spectral efficiency of (32.5 GHz$\times$8 400 Gb/s $\tilde{=}$) 1.54 b/s/Hz.

5. Conclusion

The development of future DCI technologies, which seek to exploit DWDM networking for increased spectral efficiency, may more easily facilitated in the C-band - indicating a potential departure from current GbE O-band technologies. The deployment of such high-capacity DCIs relies on the development of multi-wavelength sources which facilitate the reduction of transceiver foot-print, cost and power consumption. The results ascertained through this work clearly show how a Q-Dash MLL can contribute to the delivery of these key criteria, within the context of the DWDM ecosystem. The Q-Dash MLL wavelength channels exhibit a RIN profile sufficient to support 28 Gbaud PAM-4 transmission across 36 modes spanning $>$1 THz without additional processing/spectral shaping at the transmitter. The device is successfully implemented in an 8 channel 400G DWDM test-bed with acceptable BERs received for all channels and achieving a spectral efficiency of 1.54 b/s/Hz. Compared to other multi-wavelength sources, the Q-Dash MLL’s relatively simple single-section/single bias current design allows for the development of integrated and scalable DCI transceiver solutions, which will be critical to the continued evolution of terabit-scale interconnects.