40Gb/sec - The Next Step
How fast can you go?
Recently, there has been a flurry of announcements from vendors supporting 40 Gb/sec (OC-768), per wavelength, not just 4 wavelengths at 10 Gb/sec. The announcements started a year ago with component and router vendors and now are coming from the more traditional telecom vendors. In one sense we shouldn’t be surprised. Just as with Moore’s Law in computing, over the years we have witnessed an unrelenting drive to higher and higher transmission speeds. However, in another sense, this development is truly amazing.
Back in the optics boom time of 2000, the technology migration that people most worried about was the transition from 2.5 Gb/sec (OC-48) to 10 Gb/sec (OC-192). Most of the deployed DWDM (multiple wavelengths on a single fiber) systems then used 2.5 Gb/sec transponders with a growing number using 10 Gb/sec. Only the most bleeding-edge of the bleeding-edge startups were addressing the issues of 40 Gb/sec wavelengths. The list of issues included new sources of signal distortion which emerge at the higher speed such as Polarization Mode Dispersion (PMD) and the need for higher optical powers to drive the same distances or shorter spans.
The need for higher power can be seen when you consider that optical receivers are essentially photon counters. To them a one bit is a certain number of photons in a very short amount of time (bit time). Too few photons and the bit is seen as a zero. As the data rate increases, this bit time decreases. Thus, more photons have to be received per second. In other words, more optical power has to be used.
For DWDM systems, however, the more fundamental question is one of spectral efficiency, or the density of signals in a given amount of spectrum.
One way to view DWDM transmission over fiber is as a very wide freeway. Each lightstream is given its own lane which is a fixed width. In today’s DWDM systems, the lanes are between 25 GHz and 100 GHz wide. The narrower the lanes, the more lanes of traffic in the fiber. Today, long haul systems have typically 80 lanes. All traffic must stay within the limits of the correct lane. Within the lane, differing loads may be carried. A motorcycle doesn’t use up much of the lane, but it doesn’t carry much load. This might correspond to a gigabit Ethernet lightstream. To improve efficiency, two separate streams might be multiplexed to share a lane. On the other hand, a semi-trailer carries a large load, but hogs most of the lane.
We could increase the width of the lanes, but the total width of the road is fixed, so if we increase the width of the lanes, we have fewer lanes. This leads to the fundamental question of how many Gb/sec can be carried in a 50 GHz lane. In other words, how efficiently does the data encoding use the available spectrum (bits/Hz).
Figure 1 – Slimming down the 40 Gb/sec road hog.
Using traditional encoding schemes, 40 Gb/sec data just won’t fit into 50 GHz lanes. Back in 2000, many thought that there would soon be a fundamental tradeoff between the data rate and the number of wavelengths in the fiber, a zero sum game. If so, this might doom 40G. If it requires 100 GHz wavelength spacing to carry 40 Gb/sec traffic, why not just use 4 wavelengths spaced 25 GHz apart with 10 Gb/sec data?
Part of the reason 40 Gb/sec is now feasible is new developments in encoding, such as DQPSK (differential quadrature phase shift keying), which achieve very high spectral efficiencies. These new encoding methods have postponed the fundamental limits of physics and made the next step possible. While there were fewer than a hundred 40 Gb/sec ports deployed in 2006, there will be thousands by 2008. We will then move on to the next next step, 100 Gb/sec per wavelength.
Fortunately, creating, monitoring and protecting these new faster wavelengths does not depend on the development of new switching technology. Last year, at Interop in Tokyo, Intelligent Optical Switches, which are now used for 10 Gb/sec wavelengths, were demonstrated protecting 40 Gb/sec wavelengths. At least this migration will be easy.
