By Ambroise Thirion, Product Solutions Specialist, AddOn

Whether a consumer pull or industrial push, the number of connected devices continues to grow, each aiming to provide high-resolution video on demand (VoD) and 5G connectivity, yet at lower power consumption. In addition, the pressure of having these devices function seamlessly is felt across many industries, including telecoms, where upgrading and future-proofing networks is an ongoing process.

200Gbit/s networking for long-haul applications and 400Gbit/s networking for shorter reach, metro or data centre interconnect distances allow the greater bandwidths and higher bitrates needed for these new applications. Benefits of moving to 400G Ethernet include twice the price performance and power efficiency per bandwidth with four times greater density, and the ability to scale existing networks by 4X with the same network topology. The 400G switches also allow four-fold increase in buffer memory per chip.

Effective solution

Moving toward higher line rates is an effective solution. A recent report by London-based market insight provider IHS Markit shows a trend of 100G coherent ports shifting toward 200G and 400G, with 200G emerging as the fastest-growing segment, expected to remain strong until year 2023.

Since both 100G and 200G operate inside a 50GHz grid and do not require wider WDM grids, 200G upgrades are considered as more cost-effective.

“We forecast that 100G line-side ports will continue to decline slowly as new deployments favour use of multi-rate 200G and 400G solutions,” states IHS Markit in a recent report. “200G upgrades are easier and less expensive than upgrading to 400G wavelengths. 200G is compatible with fixed 50GHz ITU grids, used in most WDM networks deployed worldwide today.”

400G

400G is sometimes used to describe a solution that offers 400G capacity via two 200G wavelengths using traditional 32-35Gbaud technology, also defined as “dual-carrier 400G” or “dual-wavelength 400G”. 
400G is also used to describe next-generation coherent optical solutions that support a range of capacities, which include 400G on a single wavelength (e.g. “single-carrier 400G”). This is the most common meaning of 400G and is similar to how “200G” is used to describe 200G-wavelength capable solutions and “100G” to describe 100G-wavelength solutions.

A subset of the 400G coherent category is “400G ZR”, also dubbed “400ZR”, for the OIF Implementation Agreement it refers to. 400G ZR refers to a low cost, low power, coherent DWDM pluggable that only supports 400G wavelength capacity and can extend up to 120km. 400ZR solutions are expected to become commercially-available this year. 

QSFP28 4x25Gbps
The QSFP28 100Gbps is a parallel optic, 4x 25Gbps transceiver module. It has four lasers and four receivers, valid for all optical versions (100GBASE-SR4, CWDM4, LR4, ER4L).

The electrical interface is 4x25Gbps NRZ (CAUI-4) and the optical side is also 4x 25Gbps with multi-fibre push-on (MPO) connector (for 100GBASE-SR4 and PSM4) or LC connector (for 100GBASE-CWDM4, LR4, ER4L), which is a small form-factor connector that uses a 1.25mm ferrule.

With maximum power consumption of 3.5W (standardized by the MSA, or multi-source agreement), the QSFP28 100GBASE-SR4, CWDM4 and LR4 are mature products and largely used on the access side in data centres and service provider networks. Whilst the cost for the CWDM4 has dropped due to the increasing worldwide volume, it is at its lowest. However, dropping the cost requires a significantly updated QSFP28 design.

There’s a new QSFP28 100G module available, offering new a form-factor and updated laser configurations. The QSFP28 100GBASE-FR, also called QSFP28 100G single lambda, will be a single laser and a single receiver of 100Gbps instead of being 4x25Gbps (4x lasers and 4x receivers).

QSFP28 100G single laser

How is the 400G transceiver architecture helping the QSFP28 100G single laser?
400G QSFP56-DD (double-density) and OSFP (Octal Small Form-factor Pluggable, a new module and interconnect system supporting 400G optical data) are coming with three main upgrades:

• PAM4 modulation, doubling the bitrate to 50Gbps/lane;
• 50Gbauds rates, reaching a bandwidth of 100Gbps/lane (50Gbauds PAM4);
• Gearbox and PAM4 modulator embedded in a DSP.

Considering that the QSFP28 electrical host-interface is always 4x25Gbps NRZ, different alternatives are now available to “re-design” the QSFP28 with building blocks, as shown in Figure 1.

Figure 1: The QSFP28 can be re-designed with building blocks

These building blocks are:

NRZ to PAM4 electrical
The gearbox and PAM4 modulator convert four lanes of 25Gbps NRZ to two lanes of 50Gbps PAM4, remaining an electrical conversion.

50Gbps to 100Gbps
The gearbox converts two lanes of 50Gbps PAM4 to a single lane of 100Gbps PAM4. Whilst it remains electrical, the key difference is the baud rate which is converted from 25Gbauds to 50Gbauds.

Laser multiplexing
With two lasers of 50Gbps PAM4 each, these can be multiplexed over a super-channel DWDM 100GHz, as an example.

Single lambda or a super-channel QSFP28
By assembling these building blocks, two architectures are feasible:

• The single-laser 100Gbps PAM4, also called “single lambda” or 100GBASE-FR (Figure 2); and
• A super-channel transceiver with two lasers of 50Gbps PAM4 (Figure 3).

Figure 2: Single-laser 100Gbps PAM4, also known as single lambda or 100GBASE-FR

Whilst the QSFP28 electrical interface remains of four lanes of 25Gbps NRZ (CAUI-4), the signal is converted to two lanes of 50Gbps PAM4, then to a single lane of 100Gbps PAM4. The two-step conversion is handled by a DSP.

With its single 100Gbps PAM4 lane, the module requires only one laser, drastically reducing the complexity of the optical part (with one laser instead of four and no filter required). However, the DSP brings another complexity of power consumption and cost. Because of the high chromatic dispersion, this module is only intended for short reach, up to 2km.

Figure 3: Super channel transceiver with two lasers of 50Gbps PAM4

Super-channel transceiver with two lasers of 50Gbps PAM4

Another way to stay at a baud rate of 25Gbauds (and reduce the chromatic dispersion, etc.) is to mux 2x DWDM 50GHz lasers and make a super-channel 100GHz, compatible with regular 100GHz DWDM mux/demux; see Figure 4.

Figure 4: Multiplying 2x DWDM 50GHz lasers

The module has a DSP converting four 25Gbps-NRZ channels to two 50Gbps-PAM4 channels on the electrical side. On the optical side, it has two lasers of 50Gbps PAM4 multiplexed together to make a super-channel laser, compatible with regular 100GHz DWDM mux and demux. This offers the advantage of working at a baud rate of 25Gbauds, although it remains a costly option that requires a DSP and two lasers, which have high power consumption and inherent lack of reach.

Proving too expensive for short-reach uses, it’s still an effective module for upgrading from 10Gbps DWDM to 100G DWDM without changing the passive infrastructure – although active equipment will be required that may include Erbium-Doped Fiber Amplifiers (EDFAs), Dispersion compensation modules (DCMs), and others.

QSFP28 CWDM4 competition
One main application for the QSFP28 CWDM4 is in data centre, leaf-spine architecture, a two-layer network topology composed of ‘leaf’ and ‘spine’ switches. In this configuration, there’ll be a dual-fibre module with duplex LC connector, with a reach to 2km. The main cost for this module then lies with the four directly-modulated lasers (DMLs) and its filter assembly.

The super-channel transceiver (with two lasers) is not the right candidate, since it requires a DSP converting NRZ to PAM4, plus two EML lasers. Instead, this type transceiver is intended for 100GHz DWDM application, for Data Centre Interconnect (DCI).

The single-laser QSFP28 (100GBASE-FR) requires a DSP and just one laser, removing the need for a filter, drastically reducing the complexity of transceiver assembly. This reduces the cost of assembly and the components, but the DSP still remains costly. This also means that price erosion of DSPs remains crucial in the potential swap from CWDM4 to 100GBASE-FR in these applications. Today, the cost of using QSFP28 100GBASE-FR is nearly double the cost of using 100GBASE-CWDM4.

Applications
In the classical leaf-spine architecture, it is common to have servers connected to the switch with breakout topology: 4x10G or 4x25G. With QSFP28 100GBASE-FR and QSFP-DD 400GBASE-DR4, four servers can be connected at 100Gbps on a single 400Gbps port at the top-of-rack switch side.

QSFP-DD 400GBASE-FR4 is the equivalent of QSFP+/QSFP28 PSM4, whilst QSFP28 100GBASE-FR is the equivalent of SFP+ IR/LR, SFP28 IR/LR, a completely different technology but same application.