Overcoming PMD will enable 40 Gbit/s optical networks

Data rates in fiber-optic networks have increased over a thousandfold during the last decade, from about 1 Gbit/second to today's systems boasting capacity in excess of 1 Tbit/s. This increase in fiber data capacity has been achieved by alternately increasing the bit rate on the laser transmitter — time-division multiplexing, or TDM — or by adding multiple laser wavelength channels — wavelength-division multiplexing, or WDM. State-of-the art optical-transport systems carry up to 160 wavelengths at 10 Gbits/s each, for a net of 1.6-Tbit/s capacity. The next increase in fiber-optic bit rate anticipated will be raising the TDM rate from 10 Gbits/s to 40 Gbits/s, or 40G.

As the fiber bit rate has increased, optical-transport impairments within the fiber needed to be overcome. At 40 Gbits/s, the most dramatic of these impairments is polarization mode dispersion, or PMD.

A dynamically varying effect, PMD changes over time, with wavelength, and accumulates over the length of a fiber link. These complex characteristics have in the past caused many system designers to ignore PMD. But PMD can no longer be ignored. At 40 Gbits/s, as much as 90 percent of the long-haul fiber in the ground will require PMD to transmit quality signals, according to many estimates.

PMD is caused by asymmetric distortions to the fiber from a perfect cylindrical geometry. This asymmetry, or birefringence, causes different polarizations of the light to travel down the fiber at different speeds, arriving at the receiver at slightly different times. This differential delay for different polarizations of light is known as PMD. Fiber asymmetries can be caused in the fabrication of the fiber or by extrinsically induced bends and stresses caused during cabling, deployment and splicing. In addition, optical components in the system contribute to PMD.

The PMD contributions from every asymmetric stretch of fiber, every stress point, every splice and component combine to determine the total PMD of the fiber link. Some of these PMD contributions add together while others will subtract from the total PMD of the link. The PMD for the link does not grow linearly with fiber length, but scales as the square root of the fiber length.

The fiber plant changes because of such factors as variations in temperature, environmental fluctuations like vibrations from, for example, passing trains on underground fiber or weather on aerial deployments. The combination of these individual PMD contributions will change the total PMD of the link. PMD is therefore a dynamic effect, changing in magnitude the polarization angles of the PMD modes and with different magnitudes for different wavelengths. Unlike chromatic dispersion that is analytically known and (nominally) a static property of the fiber, PMD is purely statistical.

PMD causes degradation of optical signal quality. When the PMD is low, the receiver can identify ones and zeros correctly. As the PMD grows, the received signal becomes distorted and the bits spread out, interfering with neighboring bit slots. In the field, PMD events will cause network operators to notice a channel signal degradation, errored seconds, even a Sonet switch. This is confusing because all the network elements in the path will be functioning perfectly. PMD shows up as a "ghost" coming and going almost at will, and moving from wavelength channel to wavelength channel.

In a practical sense, PMD is dominated by two effects: first-order PMD, also known as differential-group delay (DGD); and second-order PMD (SOPMD), which has two components known as depolarization and polarization-dependent chromatic dispersion. Depolarization significantly dominates PDCD in real fiber networks.

At Yafo Networks we have used Monte Carlo simulations to calculate the PMD of over one billion "virtual" fibers, and have plotted the joint probability distribution function (JPDF) for various first- and second-order PMD states. Conveniently, we have found that the JPDF of PMD is universal for all fiber plants. Fifty percent of the time the PMD states of the fiber are found inside the innermost contour. The outermost contour captures the PMD states of the fiber 99.99 percent of the time.

Optical-networking transmitters and receivers (TxRx) have not been specified to operate over the entire PMD JPDF. At best, 10G TxRx have been tested and characterized by their operation due to DGD impairment only. This usually underestimates a TxRx pair's PMD performance in a real fiber environment because the significant SOPMD contribution has been neglected. Moreover, our engineers have found that different receiver components and designs can perform in significantly different ways. Specifically, we have found certain receiver designs to be particularly robust to PMD.

Standards for testing and measurement of the PMD resilience of optical-networking equipment are required as networks migrate to 40G. At Yafo Networks, we have built a PMD source that takes the statistics out of PMD, and allows for virtually any DGD and SOPMD states on the JPDF to be generated for quantifiable performance testing. Now the bit-error rate of a TxRx pair can be measured at various points on the JPDF. Since the probability of each measured point is known from the JPDF, a total outage probability can be calculated.

When the PMD of a fiber span is too great, there are several ways to mitigate the problem:

- Alternative 1. Put in new low-PMD fiber;

- Alternative 2. Subdivide the span into shorter and tolerable PMD sections and regenerate the signal more often;

- Alternative 3. Reroute the traffic through an alternative path; or

- Alternative 4. Use PMD compensators.

Deploying new fiber is quite costly and not very practical or cost-effective in the present capital expenditure-constrained economic environment. Carriers must do more with what they have. Alternatives 2 or 3 require using more equipment to transport the optical signal. In fact, the great success of optical networking over the past few years has been that it has pulled equipment, specifically costly optical-to-electrical-to-optical conversion equipment, out of the network.

Essentially, alternative 4, compensating for PMD, is the most cost-effective way to correct for PMD. A PMD compensator is a device that sits in front of, or is integrated into, the receiver, and undoes the distortions caused by the fiber plant. A good PMD compensator "tricks" the receiver into thinking the fiber is perfect.

As bandwidth demands continue to grow, networks will be pushed to ever-higher bit rates on the fiber. At 10G and even more pervasively at 40G, PMD becomes a limiting impairment to raising the bit rate. PMD cannot be predicted temporally or in wavelength: only the statistics are known. PMD varies in time due to mechanical movements to the fiber. The industry needs to adapt to this new and challenging phenomenon. Fiber and component designs need to be revisited to account for, and specify for, the impact of PMD. Testing standards for optical-transport equipment need to be updated to include PMD system tolerance. Remedies for PMD mitigation are now entering the market.