Introduction: The Challenge of Low-Frequency Noise
As data centers strive to meet ever-increasing bandwidth demands, 4-level Pulse Amplitude Modulation (PAM4) has become a critical signaling technique, doubling the data rate per symbol compared to traditional non-return-to-zero (NRZ) formats. To scale these systems efficiently, researchers are looking toward Quantum Dash (Q-Dash) Passively Mode-Locked Lasers (PMLLs). These devices act as "comb" sources, generating dozens of stable optical carriers from a single laser chip, which could significantly reduce the complexity and power consumption of multi-channel transceivers.
However, Q-Dash lasers are plagued by high Relative Intensity Noise (RIN) at low frequencies (typically below 4 GHz), primarily due to Mode Partition Noise (MPN). Because PAM4 relies on four distinct amplitude levels, it is exceptionally sensitive to such fluctuations. This noise often pushes the Bit Error Rate (BER) above the Forward Error Correction (FEC) threshold, rendering these efficient lasers unusable for high-speed interconnects without mitigation.
Methodology: Digital Line Coding as a Solution
Instead of employing complex and expensive optical hardware to stabilize the laser, this study explores the use of digital line coding to mitigate noise. The strategy is to reshape the signal's power spectrum to avoid the high-noise region of the laser. Two primary encoding schemes were evaluated:
- 8B/10B Encoding: A standard telecommunications code that maps 8-bit data to 10-bit symbols, ensuring DC balance and creating a "spectral notch" at very low frequencies.
- Manchester Encoding: A more aggressive scheme where every bit contains a transition, effectively shifting the signal's energy further away from DC and creating a wider low-frequency notch.
The experimental setup utilized a 33.6 GHz Q-Dash PMLL producing approximately 40 optical carriers. A 28 GBaud PAM4 signal was transmitted over 3 km of standard single-mode fiber (SSMF) and analyzed using a real-time oscilloscope for offline digital signal processing (DSP).
Key Insights: The Spectral Notch Strategy
The effectiveness of these codes lies in their ability to "hide" the data signal in frequency bands where the laser is relatively quiet. As shown in the power spectral density measurements, both 8B/10B and Manchester encoding suppress the signal power at low frequencies. This reduces the overlap between the PAM4 signal and the laser's high-intensity RIN, thereby improving the signal-to-noise ratio (SNR) at the receiver.
While these codes introduce overhead—20% for 8B/10B and 50% for Manchester—the trade-off is often necessary to achieve reliable transmission. For many high-density interconnect applications, the ability to use a single comb source for 40+ channels outweighs the loss in per-channel spectral efficiency.
Experimental Results and Performance
The study benchmarked the Q-Dash laser performance against a high-end, low-noise External Cavity Laser (ECL).
- Uncoded PAM4: Most channels from the Q-Dash comb performed poorly, with BERs frequently exceeding the 7% FEC limit ($3.8 \times 10^{-3}$).
- 8B/10B PAM4: The introduction of the spectral notch allowed the majority of the 40 channels to move safely below the FEC threshold.
- Manchester PAM4: This scheme provided an order-of-magnitude improvement in BER. All channels performed exceptionally well, showing only a ~2 dB power penalty compared to the "gold standard" ECL benchmark.
Conclusion and Future Outlook
This research demonstrates that digital line coding is a viable and cost-effective method for mitigating the inherent noise of Q-Dash mode-locked lasers. By shifting the signal spectrum away from low-frequency RIN, reliable PAM4 transmission can be achieved across dozens of optical carriers.
Future work will focus on developing more rate-efficient codes that maintain sufficient low-frequency suppression while reducing overhead. Additionally, implementing these encoding steps in real-time CMOS hardware (ASICs or FPGAs) will be a critical step toward commercializing Q-Dash-based optical transceivers for next-generation data centers.
Summary updated