Next-generation 40-Gbps networks are finally here.
We tend to take speed for granted, especially that of information. Ever higher rates of data transmission are fully expected, and with the petabyte becoming common , our need for faster networks clearly remains at hand.
Accommodating this demand, however, requires flexibility and expandability in addition to speed. This attitude is reflected in predictions by network developers: 10 Gigabytes per second (Gbps) is today, 40 Gbps is tomorrow, and we should get ready for 100 Gbps a few days later.
For Nortel Networks, Toronto, Ontario, “tomorrow” arrived earlier this year, when they deployed the world’s first 40 Gbps network in Europe. Roger Carroll, director of broadband terminal and modem development for Nortel Networks, says demand between data aggregation centers has been pushing this development.
“We make equipment that the average user would go to for Internet traffic. Typically this is volume aggregated from megabits per second to gigabits per second. Aggregation locations are connected to each other in a city-size type of environment. Over time the needs for communication between the central office locations has been growing,” says Carroll. “Over two decades we have gone from 2.5 Gbps for fiber to a standard that sustains many channels on a single fiber. This has opened the possibility for speeds of 40 Gbps and 100 Gbps.”
Two key factors work together for this high-speed equation: highly flexible (in wavelength) multiplexers at junction sites and effective compensation measurement for dispersion over long-distance trunks.
Dispersion comes in two forms (chromatic, CD, and the related polarization modulation, PMD), and is a phase velocity phenomenon that cause signals to degrade over distance.
Traditionally, these problems are solved through dispersion compensation modules (DCMs), which, when combined with amplification measures, keep optical signal at an acceptable level. But a lot of amplification means a lot of expense for the builder.
The advent of coherent detection, however, ushered in reconfigurable optical add/drop multiplexers (ROADMs), which use tunable lasers and filters to compensate for dispersion and maintain flexibility. A coherent detector reads both an optical electrical field and an intensity field, detecting the square of the field.
“When you square the field you introduce a large non-linearity at the receiver. If you detect the electric field you can then equalize that signal. Most of the impairments that you see are due to linear transmission. Since the linear signal is proportional to the equalized signal you have the ability to filter it,” says Carroll.
“That’s one of the strengths of coherent transmission, and Nortel was one of the first to commercialize it. In the optical domain there is no other way,” he says.
In 2005, Nortel’s Dynamically Compensating 10G Optics (eDCO) eliminated these costs by removing DCMs and amps from the equation. The coherent detection transceiver gave added flexibility to networks once operators no longer needed to redraw dispersion maps to reconfigure.
eDCO relies on tunable lasers and requires a local oscillator that can be tuned to incoming wavelength. It is basically a tunable, optical oscillator that has a reference cavity. It operates in the infrared carrier realm and is tunable in that range. The digital signal processing core is a CMOS scaled down to the 90-nm standard.
Roadblocks to 40 Gbps and higher
Moving to 40 Gbps with a traditional network implies four times the baud rate (bit interval reduced by factor of 4 from 100 picosecs to 25 picosecs). This, of course, means four times less light per bit entering a receiver, resulting in a sizable optical-signal-to-noise ratio (OSNR) of 6 db. It also exacts heavy degradation of CD sensitivity and PMD tolerance.
“Traditionally, people increased throughput by increasing the transmission rate of equipment. If we wanted to put more bits in a particular wavelength, the switch turns on and off faster. The digital intensity of the signals gets detected at the receiver. That very fast switching on and off gets distorted and you get chromatic dispersion,” says Carroll.
That hasn't prevented companies from beyond 10 Gbps, Carroll adds.
“Some have maintained this, but the pay-cost of reach they can service becomes much smaller.”
Developing a solution using eDCO
Proposals for improving traditional time-division multiplexing (TDM) networks include:
•Duobinary modulation, which changes the phase of the optical signal of each "1" bit for a certain sequence of bit and reduces the average optical signal power by one-half. Chromatic dispersion improves, but overall performance suffers;
•Differential phase shift keying (DPSK) codes the bit information into the phase of the optical signal, providing a 3 dB higher OSNR sensitivity than duobinary modulation, but receiver costs rise;
•Differential quadrature phase shift keying (DQPSK) is the four-level version of DPSK, in which each symbol transmitted has two encoded bits (00, 01, 11 and 10). Components operate at half-frequency compared to duobinary, but because it operates at 20 Gbaud it introduces complex and costly componentry.
All three solutions offer speed at the expense of CD and PMD tolerance. Also, they all limit the number of ROADMs that can be used before regeneration is required, which means service providers wouldn't be able to re-use existing architecture.
The 40 Gbps solution: QPSK
The specific method of QPSK employed by Nortel is called dual-polarization (2-POL) QPSK, because two QPSK signals are used, each modulating one of the two orthogonal polarizations of a single optical carrier. This doubling up of signals creates 4 bits/symbol, which at 10 Gbaud yields 40 Gbps.
The coherent receiver, which mixes the signal with a local oscillator, preserves the necessary characteristics of the signal to allow QPSK decoding. During the analog to digital conversion, signals are tracked by eDCO's CMOS processor, and the digital signal processor itself, developed by Nortel, compensates for CD up to +/- 50,000 picosecs/nm.
Although TDM and DQPSK suffer greatly from the effects of PMD, the 2-POL QPSK approach handles PMD even better than 10 Gbps TDM network.
Finally, 2-POL QPSK solution has kept equipment costs down, which “has allowed us to leverage 10-gigabaud component supply. We are not compelled to use high-cost bandwidth test information,” says Carroll.
This approach of using coherent receivers and comprehensive equalization mark a step forward in channel spectral efficiency, and means line equipment reduction (fewer amps, compensators, o-demuxes) and the reclamation of PMD compromised fiber plant.
So far only three customers of Nortel’s 40 Gbps technology have gone public, all in Europe, according to Helen Xenos, Metro Ethernet marketing manager, Nortel Networks.
“We anticipate customers in the U.S. as well. The interest is definitely there in all regions globally,” she says.
