Cisco Blogs
Share

PAM4 for 400G Optical Interfaces and Beyond (Part 1)

- May 10, 2018 - 0 Comments

Written by Zhenbo Xu, Technical Marketing Engineer, Transceiver Modules Group, Cisco

 

 

Non-Return to Zero (NRZ), an intuitive and simple modulation format, has been adopted for decades for both electrical and optical data transmission. However, as serial data rates increase to 25G and beyond, high-speed data lines on current printed circuit board (PCB) technology encounter significant signal loss and inter-symbol interference. This is primarily due to high-frequency roll-off during transmission along the PCB traces, power regulators and various inductances in both the pluggable transceiver and the host port. More importantly, it requires an ultrafast electrical driver and wideband distributed feedback (DFB) laser to compensate, making 50G NRZ impractical today.

In order to avoid costly electrical and optical design, there has been a recent revival of research on coherent technology and multi-level modulation formats to achieve greater than 25G channel rates. 4-level pulse amplitude modulation (PAM4), a relatively low-cost solution, has been adopted in the transceiver industry, enabling high-speed data rates, toward 400G and beyond.

How PAM4 works and why it relieves bandwidth requirements

Instead of sending binary 1’s and 0’s, PAM4 presents 4 levels, which are generated by combing two bits into a single symbol with four different amplitude levels. Each level corresponds to a permutation of the two original bits. A common way to implement this is to apply a gray coding technique to the original binary data. Decoding is simply the reverse process.

By transmitting two bits in one symbol slot, PAM4 cuts the signal bandwidth half. With half of the bandwidth, 50Gb/s data rate transmission is possible in the 25Gb/s electrical tolerance environment. Another way to look at PAM4 is that it minimizes signal degradation and doubles the data rate.

 

Challenges of PAM4 transmitter and receiver design

You may be wondering why we don’t use PAM8, or even PAM 16, to get even narrower signal bandwidth with equivalent capacity throughput. The reason is that there are both practical limitations and fundamental trade-offs for PAM-M implementation.

Multi-level signals require additional complexity in electrical signal coding, de-emphasis techniques, and characterization of transmitter optical signals which can be subject to transition-induced misalignment of the multilayer eye diagram.  And receivers need to deal with electrical clock recovery, digital filter equalization and decoding, which require additional development effort and circuit design cost. All of these drive the transceiver industry to innovate technologies and signal measurement techniques.

Another challenge with more levels is related to noise. Shannon’s fundamental information theory imposes a ceiling on maximum line speed vs. available SNR. PAM-M sacrifices SNR for data rate. For example, the equivalent of an eye diagram for PAM4 consists of three eyes stacked on top of each other, meaning that the signal-to-noise ratio (SNR) for each eye is less than one third that of NRZ.

In future posts of this blog series, we’ll get more into detail about the impact of PAM4 on the optical transceiver industry.

In an effort to keep conversations fresh, Cisco Blogs closes comments after 60 days. Please visit the Cisco Blogs hub page for the latest content.