Optical communication systems face a shared challenge at opposite ends of the distance scale: carrying more information while working within strict energy budgets. Data center links must balance throughput with transceiver power, size and cost. Long-haul submarine cables face limits on the electrical power available to undersea amplifiers, which constrains optical power and throughput per fiber. Addressing these needs calls for different approaches to modulation, receiver architecture and channel power allocation.
Why data center links need new design choices
Data centers have traditionally increased link capacity by adding fibers or wavelengths carrying on-off keying (OOK). As demand rises, scaling that approach can increase the number of optical lanes and the associated component count, power consumption and cost. A key objective for next-generation links is therefore to carry more data per wavelength while maintaining practical power margins and tolerating losses in the fiber plant.
Fiber attenuation reduces received signal power, while chromatic dispersion (CD) can distort intensity-modulated, direct-detection signals. In these links, information is carried in optical intensity rather than in the full electric field. Dispersion and transmitter chirp can combine to create frequency-dependent fading, including response notches that make receiver equalization more difficult. The effects constrain achievable reach and data rate, particularly as symbol rates increase.
Direct detection: practical formats and sensitivity
Direct-detection (DD) systems are attractive for short-reach applications because they avoid some of the complexity associated with coherent detection. However, they use only the intensity degree of freedom of the optical signal, placing limits on how far their performance can scale.
Among the DD-compatible formats assessed for links beyond OOK, four-level pulse-amplitude modulation (4-PAM) offers a useful balance. It has relatively low implementation complexity and showed better tolerance to noise and distortion than orthogonal frequency-division multiplexing (OFDM) in the comparisons described in the dissertation. The format was subsequently adopted in IEEE 802.3bs approaches to 400 Gbit/s using multiple transceiver lanes.
Receiver sensitivity is another concern: a receiver needs sufficient optical signal power to distinguish symbol levels reliably. Avalanche photodiodes (APDs) and semiconductor optical amplifiers (SOAs) can improve sensitivity for direct-detection systems. The reported analysis found improvements of 4 to 6 dB, although practical designs must account for device behavior and system constraints. Such measures can extend the usefulness of DD formats, but they do not remove the underlying challenge of scaling them beyond 100 Gbit/s per wavelength.
Coherent detection without conventional DSP overhead
Coherent receivers can access four signal dimensions: two quadratures across two polarizations. This broader use of the optical channel can support higher spectral efficiency and improved noise tolerance. Conventional coherent implementations, however, rely on high-speed analog-to-digital converters (ADCs) and digital signal processing (DSP), which can be difficult to accommodate within data center power and cost budgets.
DSP-free coherent receiver architectures offer a different design path. The proposed approaches use analog signal processing to avoid high-speed ADCs and DSP while aiming for performance comparable to DSP-based counterparts. For engineers evaluating these architectures, the practical question is not simply whether coherent detection offers more signal dimensions. It is whether the receiver can recover polarization and carrier information with acceptable complexity and power for the intended link.
This distinction matters in system planning. Direct detection may remain suitable where reach, rate and power margins fit the application. Where those constraints become difficult to meet, coherent approaches can provide additional channel degrees of freedom, provided their implementation overhead can be controlled.
Submarine links: allocating scarce amplifier power
Undersea cables face a different energy constraint. Repeaters spaced along the route require electrical power delivered from shore, and only a portion of that power is available for optical amplification. The amplifier budget therefore limits the optical power that can be launched into each fiber. At the same time, throughput is affected by amplifier noise and Kerr nonlinearity, which becomes relevant as optical power increases.
Spatial-division multiplexing (SDM)—using multiple spatial dimensions such as modes, cores or fibers—can help address energy limits by distributing information across more dimensions at lower power per dimension. But the number of dimensions is only part of the design problem. The power assigned to each wavelength channel also affects the capacity that the system can achieve.
The submarine-link analysis formulates channel power allocation as a capacity optimization under an electrical power constraint. It accounts for amplifier physics, noise, Kerr nonlinearity and power-feed limits rather than treating the available optical power as independent of the repeater design. In the studied system, optimized channel allocation increased capacity by about 70% compared with the theoretical capacity of a previously proposed high-capacity system. This result is tied to the modeled system and should not be treated as a universal gain for every cable.
Applying the findings to system design
These approaches suggest several practical considerations for technical teams:
- For data center links: evaluate modulation format alongside reach, dispersion, receiver sensitivity and implementation complexity.
- For DD designs: consider whether sensitivity improvements from APDs or SOAs provide enough margin for the target link, while recognizing the limits of intensity-only signaling.
- For coherent designs: assess receiver architectures that reduce reliance on power-intensive high-speed conversion and DSP.
- For submarine systems: optimize wavelength-channel power in the context of repeater power, amplifier behavior, noise and nonlinear effects.
Conclusion
Higher optical throughput is not only a matter of increasing symbol rates or adding wavelengths. Data center links must reconcile capacity with tight transceiver power and complexity budgets, while submarine systems must make the most of a limited supply of repeater power. Choosing suitable modulation and receiver architectures for short reach, and optimizing channel power under realistic energy constraints for long haul, can help engineers make better use of the available optical channel. In both settings, system-level trade-offs—not any single component or format—determine what capacity is practical.