Low Distortion All-Optical Threshold Device

This invention relates to all optical signal processing devices.

Optical fibers have a signal carrying bandwidth that potentially is in the multi-terabit range. This makes optical fibers an attractive choice for signal transmission systems, particularly in long haul transmission systems where the cost of the cable makes up a large fraction of the cost of the communications link. Fiber is far superior to any competing technology in terms of bandwidth, cable size and other factors. However, signal attenuation in long fiber segments is still an issue that has to be addressed. To compensate for signal losses, current optical fiber transmission systems employ electronic repeaters at spacings on the order of tens of kilometers. To detect, re-synchronize, and regenerate the signal, the repeater must convert the optical signal to electrical form, amplify it, and reconvert the amplified signal to optical form. Aside from the complexity and expense associated with such conversions, the need to work in the electronic domain is limiting the bandwidth of the overall transmission system.

It is expected that electronic regeneration will be replaced by fiber amplifiers, which have an overall information-carrying capacity comparable to that of the fiber itself. Erbium fiber amplifiers represent a new technology and promise to provide better performance at a lower cost. While amplifiers can compensate for fiber attenuation by boosting the signal level, the functions of signal regeneration and signal re-timing are not addressed by the amplifier. Without these functions there is no means of restoring the data, and any noise which appears at any point in this type of system merely accumulates.

Moreover, broadband noise over the entire gain bandwidth of the amplifier (known as amplified spontaneous emission) is an inevitable by-product of the amplification process. Consequently, the amplifiers themselves are a significant source of noise in the system. Because the signal itself does not generally occupy the entire amplifier bandwidth, a spectral filter can be used to stop all noise outside the signal bandwidth. While this is an important step in reducing the bit error rate at the receiver, it does nothing to the in-band noise; and it is that noise which generally causes the most errors.

An approach for reducing the noise in the baseline of an optical signal is described in "Pulse shaping, compression and pedestal suppression employing a nonlinear-optical loop mirror", Doran et al., Optics Letters, Vol. 15 No. 22, Nov. 15, 1990, pp. 1294-1296, where the authors reported on the signal transmission properties of a sagnac interferometer. They note that the input/output transfer function 10 of such an interferometer is oscillating between peaks and troughs (see FIG. 1), and that the oscillating function is bounded by two lines: one is a 45 degree line starting at the origin (line 11), and the other is by a nearly horizontal line (line 12). They also note that by using only that portion of the input/output transfer function that starts at the trough next to zero input and ends at a relatively linear portion of the transfer function, the baseline of the signal is compressed relative to the rest of the signal. Any noise that rides on top of that baseline would be compressed.

The fact that the slope of the input/output transfer function starts at a low value and increases as the input power increases also causes the output pulse to be narrower than the input pulse. Doran et al. note that this phenomenon may be thought of as pulse shaping in the form of temporal pulse compression. Increasing the signal into the interferometer much further first shortens the tips of the pulses and generates substantial distortion of the pulses.

The transmission capabilities of optical fibers are enhanced by providing all-optical means for removing noise from optical signal pulses and for creating clean output pulses with specified characteristics. This is accomplished with a nonlinear interferometer that advantageously cooperates with an applied digital signal having a low level and a high level. More specifically, the nonlinear interferometer is designed to have a relatively level region in its input/output transfer function in the neighborhood of the low input signal and in the neighborhood of the high input signal. Depending on the input signal characteristics, it is sometimes beneficial to design the interferometer so that the nominal high level of the input signal falls approximately in the middle of its corresponding flat region. In another embodiment, a bandpass filter is included at the output of the interferometer to remove the excess bandwidth that is created by the preceding amplifiers and the interferometer itself. The filter also removes the out-of-band noise in the input signal.

FIG. 1 depicts the input/output transfer function of a nonlinear interferometer;

FIG. 2 presents a block diagram of a long haul fiber transmission system;

FIG. 3 depicts the transfer function of an ideal threshold device;

FIG. 4 illustrates one embodiment of an all-optical threshold device, comprising a nonlinear sagnac interferometer; and

FIG. 5 depicts the transfer function of a nonlinear sagnac interferometer as designed for the purposes of this invention.

As indicated above, the very large bandwidth offered by optical fiber makes it an attractive candidate for congested long transmission routes. In particular, optical transmission appears to be an attractive candidate for undersea cable applications. The challenging characteristics of such an application are the large span of the cable, and the need to place amplifiers undersea. One consequence of the latter is that the amplifiers must be extremely reliable. The advent of Erbium amplifiers has heightened interest in undersea applications, but the issue of how to maintain the integrity of pulses as they travel through the fiber and through the amplifiers was heretofore not fully solved.

This problem is solved by the all-optical low distortion threshold device of this invention, which can be employed in a multi-amplifier transmission system as depicted in FIG. 2. In such a transmission system, one or more amplifiers 15 are followed by a threshold device 20. Each amplifier 15 optically amplifies the incoming bi-level signal and after a number of amplifier stages the threshold device suppresses the noise and forms the pulses into the desired pulse shape.

FIG. 3 depicts the input/output transfer function of an ideal threshold device. When the input signal is binary with nominal levels A and B, all noise is suppressed in such a device when the superimposed noise level is not too great. The threshold device also fixes the magnitude of the output pulses (CO and C1) and, to an extent, determines the pulse shape. That is exactly what is needed for the optical signal application of FIG. 2; to wit, an all-optical device with a bandwidth commensurate with that of the transmission fiber, a device that offers zero transmission for any signal with energy below a standard threshold energy, a device that converts any pulse with energy greater than the threshold to some standard fixed energy and shape, and a device that forces the resulting pulse to have a predetermined temporal width and a predetermined spectral width.

A recently disclosed use of sagnac interferometers employs the interferometer as a switch. Such a switch is disclosed, for example, in U.S. Patent application Ser. No. 07/521774 filed May 10, 1990, still pending. When investigating the use of the sagnac interferometer as a switch, artisans have concentrated on creating an element that switches as completely and efficiently as possible; and that meant that the trough marked 13 on FIG. 1 had to be made as deep as possible. In contradistinction, exactly the opposite tack is taken for this invention. Specifically, to make the input/output transfer function approach the function of FIG. 3, it is desirable to make the trough as shallow as possible.

FIG. 4 presents a diagram of a sagnac interferometer having a 2 by 2 optical coupler 30 and a nonlinear fiber 40 connected between output ports 31 and 32 of the coupler. A signal pulse applied at port 33 is split and passed to ports 31 and 32 in relative strengths proportional to the coupling balance. The pulse exiting port 31 travels through the fiber and re-enters coupler 30 at port 32. Similarly, the pulse exiting port 32 travels through the fiber and re-enters coupler 30 at port 31. The intensity dependent Kerr effect in the fiber provides the nonlinearity in the device. With a perfect 50--50 coupler, the two counterpropagating pulses in the fiber loop receive the same phase change and the Sagnac acts as a reflector. Unbalancing the loop causes an intensity-induced differential phase between the two pulses, and that causes the Sagnac to transmit rather than to reflect. There are three ways to achieve the imbalance: (a) by a coupler that does not split the pulses evenly, (b) by asymmetrically including gain in the fiber loop, i.e., by putting the gain at one end of the fiber loop, and (c) by asymmetrically including loss in the loop. Of these, approach (a) is the one approach that creates the desired change in the shape of the transfer function. In addition, in order to develop a large difference in the power of the pulses, approaches (b) and (c) may also be used, so that the power level of the optical signal can be reduced. Approach (b) results in the addition of further noise from the amplifier in the loop so it is less preferred than approach (c). A coupler that does not split the pulses evenly can be simply purchased. A coupler that creates the imbalance by attenuation can be created by simply adding some attenuation means to one of the ports, such as at port 32. This is illustrated by the wider line segment 35 near port 32. A coupler that creates the imbalance by adding gain into the loop requires an optical amplifier. That is not depicted explicitly in FIG. 4 to avoid confusion due to the depiction of the "loss" approach. It should be understood, however, that placing an optical amplifier in some proximity to port 32, instead of the loss means, is a relatively simple task.

FIG. 5 illustrates the input/output transfer function 16 of a nonlinear sagnac interferometer that is substantially unbalanced, as desired for the purposes of this invention. It starts with a low but increasing positive slope in region 21. In region 22 the slope starts at 1, with increasing input power it first increases and then decreases back to 1. In region 23 the slope decreases to zero at the peak point 24, turns negative, increases in the negative direction to some maximum value, then decreases in magnitude till it reaches zero again at dip point 25, and then turns positive again. Regions 21 and 23 are "saturation regions" in that they resemble the saturation regions in a magnetic B-H curve.

For purposes of this invention, operation of the sagnac is restricted to regions 21,22 and that portion of region 23 that includes peak 24 and is non-dipping. The transfer function is said to be non-dipping if point 25 is not below the decision level which, for most applications, might be the midpoint of region 22. Moreover, it may be desirable to further limit the operation of the sagnac to that portion of region 23 that is characterized by a derivative that is less than one in the absolute sense. In region 21 the derivative is less than 1 by definition.

It should be noted that the above discussion of FIG. 5 and the slopes associated with the regions can also be viewed in a normalized sense rather than the absolute sense. That is, the transfer function can be normalized to the average slope in the nearly constant slope portion of region 22. When so normalized, the region 21 should be limited to normalized slopes that are significantly less than 1, such as 0.3 or smaller. The transfer function can be normalized in a different manner as well, such as by developing the function dP.sub.out /dP.sub.in divided at each point by P.sub.out /P.sub.in. The saturation regions are the areas where the quotient is less than one.

In utilizing the unbalanced nonlinear interferometer as a threshold device, it is useful sometimes to assess the expected noise levels when the optical signal is high. That is, whereas no operating point placement needs to be controlled for the low optical signals, the operating point placement for high optical signals can, and perhaps should, be controlled. This comes about from the fact that transfer function 16 has two neighborhoods in region 23 where the derivative, or slope, is very close to zero where excursions due to noise are essentially eliminated. When the expected noise level is low, the optimum operating point for high optical signals is in the neighborhood of peak point 24. On the other hand, when operating nominally at point 24 (when the signal intensity is high), large noise-induced excursions on the negative side might actually cause a detection error. When such a possibility exists, the nominal operating point is in the neighborhood of point 25.

While it is appreciated that the apparatus of FIG. 4 can be designed to exhibit a transfer function like the one shown in FIG. 5, and that such an apparatus can be utilized as a threshold device, it should also be appreciated that when used as a threshold device it tends to distort the temporal shape of pulses which pass through the device. The wings of the pulse tend to be clipped, thus shortening the pulse duration, and the center of the pulse tends to be flattened when the "high" operating point is set at point 24. Even a "dimple" can be created in the output pulse when the "high" operating point is set at point 25.

One solution to this temporal pulse distortion is to use special optical pulses at the input, known as "solitons", which can self-correct for the distortions induced by the nonlinear sagnac interferometer. However, the use of solitons, or pulses that approximate solitons, does have a practical disadvantage. In typical communications applications the pulses have a duration which is on the order of 50 picoseconds. Solitons which can be used in a nonlinear sagnac interferometer, on the other hand, are on the order of one picosecond or shorter. Generation of pulses that are this short may require a special laser.

Another approach for overcoming the pulse distortion induced in the nonlinear sagnac interferometer that is simple, inexpensive and practical and involves the use of a spectral filter 50 to recover the desired pulse shape after passage through the Sagnac. The particular type of distortion that the nonlinear sagnac interferometer imparts on the pulse envelope (both intensity and phase profile) as it passes through tends to involve errors on the high frequency components of the spectrum. Spectral filter 50, which may be a conventional commercially available Fabry-Perot filter, is connected to port 34 of coupler 30 to block the unwanted components. This filter is considerably narrower than the filter used to block noise from the amplifier, and has a passband width that is approximately equal to the spectral width of the signal itself (a "matched filter").

In an experimental setting, an unbalanced coupler with a 60-40 splitting ratio was used in combination with a filter having a bandwidth of about 80% of the bandwidth of the desired pulse. The pulses used were 60 psec wide and the fiber was approximately 500 m long.