All-optical logic revived for fiber applications

Two recent papers suggest that optical computation may be re-emerging as a solution for fast communications. Work at the California Institute of Technology (Pasadena, CA) and the University of Maryland (College Park, MD) suggests that it may eventually be practical to perform basic processes such as error correction before detection, thus reducing latency and simplifying the processing electronics. The first of these projects is specifically aimed at WDM applications, and particularly at bringing back the idea of a spectral bus. The second is the demonstration of a new system that could allow logical operations at 100 Gbit/s.

In the transmitter (Tx), three external cavity lasers (ECL) are used as signal sources that are polarization shift-key (PolSK) modulated with data from a pseudo-random pattern generator (PPG). At the receiver (Rx) the light is split in two and amplified (by erbium-doped fiber amplifiers, EDFAs). In the upper arm, four-wave mixing (FWM) only takes place when all beams are present [111]. Other cases are dealt with in the bottom arm, where the birefringent element [BE] rotates one of the three beams (marked S) by 90°. BPF: optical bandpass filter. SOA: semiconducter optical amplifier. FRL: fiber ring laser (to limit the SOA gain). O/E: opto-electronic converter. OSA, MTA: optical spectrum analyzer, microwave transition analyzer (to examine sequences in the frequency and time domain respectively).

The spectral bus, which was originally proposed by researchers at IBM in the late 1980s, is a way of using the wavelength channels within a fiber to transmit data cooperatively, in parallel, rather than in independent serial streams. The idea is to use each wavelength to represent a different bit in a large word, thus exploiting the entire bandwidth and maximizing speed for a single communication channel. Until recently, this approach had not been possible due to dispersion: bits of words sent together would arrive separately, because each wavelength travels at a different velocity through the fiber. With recent developments in dispersion compensation, however, researchers suggest it may be time to reconsider the idea.

To process the amount of incoming data the spectral bus would bring in, the Cal Tech team says that all-optical logic would be necessary, and it has developed a scheme that is compatible with the constraints of WDM. Instead of intensity modulation, its proposed system uses orthogonal polarizations of light to represent 1s and 0s: thus allowing four-wave-mixing (FWM) interactions that only take place when all relevant beams are present. The mixing itself takes place in a semiconductor optical amplifier (SOA) and, in this scheme, is nondegenerate: all wavelengths are different. Degenerate FWM, which will also occur, must be filtered out.

One possible application of this logic was demonstrated in an experimental implementation of the Hamming error-correcting code. Words, in this case, had one data bit and two parity bits, which meant that correct signals could only be [111] or [000]. In logic, the implementation involves ANDing each of the three possible pairs of bits together, and ORing the results of these and gates. This way, if any two of the bits is a 1, the output will be a 1, if not, it will be a 0. Though this particular example (see figure) may not scale well, researchers say it is the first example of an experimentally verified, single-level three-input all-optical gate. In addition, they claim an integrated implementation could run at up to 40 Gbit/s.

The University of Maryland team has focused on a different mechanism for providing optical nonlinearity in its system. It uses counter-propagating beams of pulses through an electro-absorption modulator (EAM) to implement its optical AND gate, and have tested the system on a 2³¹-1 pseudorandom bit sequence. The way the EOM implements the Boolean function is very simple. When a pulse of sufficient intensity propagates through the EAM, charge carriers are produced. These screen out the electric field across the device and reduce the absorption through it. Thus, where two pump pulses collide, their propagation is efficient and a pulse emerges: where only one pump pulse is present, absorbtion is high and that pulse is extinguished.

How fast the system can go is only limited by the recover time of the EAM gate, say researchers, which in their experiment approached 10 ps. This would allow a processing rate of 100 Gbit/s. The fact that the inputs and outputs are the same wavelength, they say, also makes their gates cascadable: this is not true for systems where the interaction of two beams is read out via a differently-colored probe. They are currently looking at ways to reduce the power needed to drive the nonlinearity, including using interferometric techniques to exploit phase modulations produced by the EAM: reported experiments used up to 5 pJ per pulse.

For information, contact Per Olof Hedekvist, now at Chalmers University Sweden, e-mail: [email protected], or Ehab Awad at the University of Maryland; e-mail: [email protected].

Sunny Bains

REFERENCES