Author Archives: Serenity Huang

100G QSFP28 PAM4 or Coherent CFP?


The ever-increasing need for higher data rate in mobile data traffic, data centers and cloud services has pushed the access streams from 2.5Gb/s to 100Gb/s, and is demanding for 100Gb/s beyond without a stop. In today’s core network that has deployed 100G rates, there are QSFP28 optical transceivers including SR4, PSM4, CWDM4, LR4, ER4, etc., serving for a maximum 25km transmission distance. And there are 100G AOC, DAC and breakout cables generally for applications of tens of meters. 100G CFP/CFP2 modules including SR10, LR4 and ER4 support transmission distances of 150m to 40km. Until recent years, the telecom service providers are adopting new 100G DWDM technologies in their high capacity and long distance backbone applications. Coherent 100G DWDM transceivers are the first to be deployed for 100G long-haul applications, and then new technologies like PAM4 (Pulse Amplitude Modulation) are developed to meet lengths requirements for 100G metro network. This post is to discuss the issues on coherent and PAM4 100G DWDM transceivers.

Overview on 100G DWDM Transceivers

In the past few years, the adoption of 100G DWDM technologies is mainly focused on coherent DWDM optical transceivers, including CFP and CFP2. Until the year 2016, Inphi (a specialist in this area) offers pluggable 100G PAM4 QSFP28 DWDM transceivers to support 80km data center interconnect (DCI). The alternative for 100G DWDM coherent transceiver is given much attention. Besides, this new option for 100G DWDM transceiver also arouses hot discussion on which to choose. Knowing the characteristics and suited applications of them could help in selection.

QSFP28 PAM4 and Coherent CFP/CFP2

There are significant differences between QSFP28 PAM4 transceivers and coherent CFP/CFP2 transceivers, but they also have some relations in 100G applications. Contents below will go to details of these optical modules.


Before the announcement of PAM4, binary NRZ (non-return to zero) modulation format is used for 40G and 100G long-haul transmission systems. PAM4 has four distinct levels to encode two bits of data, essentially doubling the bandwidth of a connection. Currently the single-wavelength PAM4 modulation scheme is considered the most cost-effective, efficient enabler of 100G and beyond in the data center. The 100G DWDM transceiver utilizing PAM4 signaling is in QSFP28 form factor. The advantage is that the customers who want to build an embedded DWDM network can use this transceiver directly in the switch. On this side, it is simple and cost-effective solution. But there are some prerequisites: it needs amplification to get out of the blocks and dispersion compensation to go beyond 5-6km. Therefore, a separate DWDM multiplexer with an amplification system and dispersion compensation is required to connect data canters together.

single wavelength PAM4 100G

In another case, if the QSFP28 PAM4 module is added to an existing DWDM network, it must be a network already having right dispersion compensation modules (DCMs) and amplification system in place; if it is not, changes are required when QSFP28 PAM4 is later added.

Coherent CFP

CFP digital coherent optics (DCO) have a high speed digital signal processing (DSP) chip built in. They do not require separate DCMs. This is what makes CFP different from QSFP28. Instead, they have electronic dispersion compensation built in. Although the built-in DSP requires more power and adds cost in components, it releases the switch vendors from adding DSPs to their equipment. Coherent CFPs enables transmission distance of more than 1000km between sites.

CFP2 analog coherent optic (ACO) is half the width of the CFP. Existing CFP2 coherent DWDM optical transceivers are analog and require a separate DSP on the host board to take the full advantages of the coherent features. So it is suited for switch vendors who have fitted such a DSP, but it adds additional cost and power consumption on the main board.

CFP2 digital coherent optic (DCO), expected to be released in the coming two years, is more optimized than CFP2 ACO in that it has built-in DSP. This component will open up to all switch vendors using CFP2 without DSP. With different coherent CFPs optional, customers can pay only for what they need when they need it.

QSFP28 PAM4 or Coherent CFP?

This really should depend on the applications. According to ACG research (an analyst and consulting firm that focuses on the service providers’ networking and the telecom industry), the 100G PAM4 solution and coherent DWDM solution, together with IEEE802.3ba, cover different portions of the optical fiber reach in the data center interconnect. So when deploying a long distance 100G DWDM network with DWDM transceivers, the required transmission distance and available equipment should be taken into consideration when choosing a suitable pluggable module.

IEEE, PAM4, OIF coherent optical reaches


Using pluggable transceivers for embedded DWDM, where the DWDM functionality is in the transceiver and not a separate DWDM converter platform, offers the ultimate solution in terms of cost and simplicity. Both QSFP28 PAM4 and coherent CFP/CFP2 are all suited to this approach. They can be used for embedded DWDM networking or as part of an existing DWDM installation. They all enable the advantages of pluggable modules: simple installation, easy spares handling, lower cost of ownership and quick return on invest.

Simplify the Implementation of High Density 100G CXP

Data center bandwidth demands continue to grow, requiring higher capacity and throughput. The 100G/120G Ethernet is no longer new in data center optic market, but it’s still a complex act to efficiently and effectively upgrade existing 10G/40G architectures to these higher data rates, especially in a space-constrained application. In order to explore the approaches of smooth migration to high density 100G/120G network, this post will take multimode 100G CXP module as example, and illustrate some simplified scenarios when upgrading to these higher data rates.

Overview on 100G CXP Module

High density 100G CXP is very popular in the implementations up to 100Gbps for saving-space. This deployment can then leverage the 10G-per-lane channels to distribute the 10G data anywhere in the data center. 100G CXP module is designed to connect with an MTP/MPO-24 connector, which can be divided to 10x10G or 12x10G transceiver pairs. For 120G CXP, it is also possible to separate the signals into three QSFP+ transceivers, and then to three groups of 4x10G transceivers by using an 8 fibers MTP/MPO to LC breakout module or cable.

Direct Connectivity for two CXPs

For two 100GBASE-SR10 modules, direct link can be easily made via 100G MPO cable. For connecting two 120G CXPs, a cost-effective 24 fibers MPO trunk can also work well. Here uses an 24 fibers MPO (female) to MPO (female) OM4 polarity B trunk cable.

direct link for two 100G/120G CXP modules

Figure 1: direct link for two 100G CXP modules.

Connectivity Methods for CXP and SFP+/QSFP+

In this part, the scenarios applied for 100G to 10G connection, and 120G to 40G or 10G connection will be explained.

100G to 10G

Figure 2 shows a direct link for one 100G CFP module and ten 10G SFP+ modules. By using the 24 fibers MPO to LC duplex harness cable, the whole 100G from the CFP transceiver is connected to ten SFP+ transceivers (two LC duplex legs are not used in this link). The fanout legs are available to be the same length or staggered type, so as to meet different applications.

direct link for a 100G CFP to 10x10G SFP+s

Figure 2: direct link for a 100G CFP to 10x10G SFP+s.

In figure 3, the interconnect for CFP and SFP+ transceivers is more flexible than the direct link. Here the 160 fibers MTP/MPO (male) breakout patch panel allows connectivity to any duplex path reachable by the patch panel. This method offers ultimate flexibility in allowing connectivity to any row, rack or shelf. Moreover, this breakout module can support up to eight groups of this 100G to 10x10G transmission. In such a high density link, it is suggested to use HD patch cables or LC uniboot patch cables to enable quicker and better cable management.

interconnect solution for 100G CFP to 10x10G SFP+s

Figure 3: interconnect solution for 100G CFP to 10x10G SFP+s.

120G to 10G and 40G

When directly connecting one 120G CXP to twelve 10G SFP+ transceivers, a 24 fibers MTP-24 to 12 LC harness cables can do the job well. Here we use a customized high density bend insensitive female MTP-24 to 12 LC duplex OM4 breakout cable.

 direct link for 120G to 12x10G transceivers

Figure 4: direct link for 120G to 12x10G transceivers.

An option for breaking out a 120G CXP to three 40G QSFP+s is to use the 1×3 MTP/MPO conversion harness cable. Figure 5 illustrates implementation of a 1×24 strand MTP to 3×8 strand MTP conversion harness cable. Like the 12x10G segregation mentioned above, once split, the 3×8-fiber QSFP+ channels can be distributed through patch panels and 12-fiber based trunking to any area of the data center.

hybrid link for 120G CXP to 40G QSFP+s and 10G SFP+s

Figure 5: hybrid link for 120G CXP to 40G QSFP+s and 10G SFP+s.


This article has illustrated some simplified implementation examples of 100G CXP modules. 24 fibers MTP/MPO trunk cable are suited for connecting two CXP modules. Breakout cables can achieve quick connection for CXP and QSFP+ or SFP+ optics, but when flexible patching is needed in the link, it would be better to adopt breakout patch panel. If you need 100G optics, FS.COM can offer you fully tested compatible 100GBASE-SR10, 100GBASE-SR4, 100GBASE-LR4 and 100GBASE-ER4 transceivers, etc.