OVER-ALL DESCRIPTION OF DESIGN AND OPERATION
A basic element in optical switching systems is the crossbar switch shown in FIG. 1 in which any input light beam, 1, on the left is delivered to an outpui, 2, on the right without interference from other beams. Various designs exist in the literature for non-blocking and blocking switches: our interest is in non-blocking in which an input at a given port may be sent to a second available or free output port with requiring any other established connection to be changed.
The inputs are normally light beams carried by fibers, The light may be in the form of either a single wavelength or may be a composition of many wavelengths as is found in Dense Wavelength Division Multiplexing (DWDM) systems. The light that enters the switch can be in many forms. The optical switch is wavelength, modulation rate and protocol independent. All can be accommodated by the switch, mixtures are also supported/The input may also be any properly collimated light beam.
The goal of the switch is to switch the input channel to the selected output channel as quickly as possible. SONET protocol requires that is less than 10 milliseconds. This switch is typically less than 3 milliseconds. The second goal of the switch is to achieve minimal loss. This switch is designed to be less than 1.5 dB insertion loss.
The light to each input port is collimated across the free space region between the input and the output port. Since the input to output ports can vary in distance, careful design considerations are taken to minimize the losses as this ratio become large for larger switch matrices. This achieved by placing the light launched from the fiber at the focal length of the lens. The size of the lens is chosen to create as flat of a wavefront as possible to minimize the variation in coupling between the input and output ports.
The basic building block of the embodiment of the present switch is a device, 3, that directs the input to the output where a similar device, 4, simultaneously directs the output to the input. This dual pointing is necessary to ensure that the maximum light power is directed from the input to the output. The basic device is replicated N times in both input and output to make an NxN switch and N and M times respectively to make an NxM switch.
Constructing the base pair of Light Directors
FIG. 2 shows the plan view of the light director while FIG. 3 shows the side elevation. The light derived from the input fiber is reflected downwards from a fixed mirror, 5, to a rotary mirror, 6, shown as a triangle or wedge from which it is transmitted to the output rotary mirror, 7, from which it is reflected upwards as shown to a second fixed mirror, 5, from which it is reflected out to form the output beam. The light beam impinging on mirror 6 is swept in a horizontal plane orthogonal to the vertical (as show axis of the rotary mirror6 to be picked up by the output rotary mirror 7.
The input light shown in FIG. 3 is generated from the fiber by a lens system. FIG. 3 shows a fiber,8, contained within a cylindrical ferrule, 9, that also contains a grin lens, 10, and a 45° mirror, 11, to reflect the light orthogonally downwards. A grin (gradient index of refraction) lens is to be preferred for space and compactness reasons. Placing the exit pupil of the fiber at the focal point of a (collimator) lens makes a normal collimated (i.e parallel) beam at the lens output. FIG. 4 also shows a 45° mirror, 11, placed after the grin lens to reflect the light orthogonally to the fiber axis as required in the side elevation drawing of FIG. 3. The exit pupil of the fiber in FIG. 4 must match the entrance pupil of the grin lens for maximally efficient light collection and propagation. This design requirement ensures that all of the light enters the lens and exits in the output beam thus minimizing the insertion loss. Low back reflection is ensured by cleaving the fiber by a few degrees (usually at 8 degrees), or if a GRIN lens is used as the collimator by polishing the face at an 8 degree angle shown schematically in FIG. 4. The resulting output beam is collimated but detailed study by Rayleigh showed that the output beam will have divergent properties determined according to the well-known Rayleigh formula that describes the necking of the beam and the location of the neck. The neck position, in normal optical practice, may be placed at the entrance pupil of the receiving lens but sometimes the mid-point of the between the input and output is used.
The collimating GRIN or lens axis must be collinear and on-axis with the fiber axis. This is usually accomplished by using ferrule with a precise, centered inner diameter in which is placed a fiber with a precise outside cladding diameter and a GRIN lens with a precise outer diameter. The ferrule must be long enough to ensure that the fiber and lens are sufficiently collinear as to avoid angular misalignment. Choosing the correctly matched inner and outer diameters and length results in an assembly that is mutually bore-sighted. During the assembly process the separation between fiber and GRIN or other lens is controlled to be the desired focal distance to place the Rayleigh waist at the correct desired distance down the beam. The parts are usually cemented in place using epoxies but the may be soldered in place id desired if metalized components are used.
As is well known using cleaved or angle polished components results in beam walk-off of a few degrees. That is the output beam will exit the GRIN at a non-zero angle to the GRIN and/or ferrule axis and in the dihedral plane defined by the polishing angle. If both fiber and GRINs are polished they must be mutually. aligned. (The assembly operated in a gross sense as a prism.) A. thin prism may be used to affect the walk-off or the mutual centers of the fiber and GRIN may be off-set by the correct amount.
We desire the beam to be orthogonal to the axis so we use a mirror or prism place in the ferrule to reflect the axial light in an orthogonal direction. Normally setting the mirror at 45 degrees to the optical axis would do this. But due to the walk-off effect, we adjust the mirror angle to 45 degrees plus or minus the walk-off angle and then we fix it in place using glue, epoxy, solder, welding, or other appropriate joining technique.
In some applications it is enough to assemble the ferrule, fiber and GRIN and epoxy it directly and as needed rotate the fiber to the correct angle as needed. Sometimes the fiber comes in an array and cannot be twisted without compromising the loss in the fiber assembly. In this case, it is necessary to assemble all the parts in the ferrule and to rotate and fix the ferrule assembly at the correct angle to the fiber as the last step.
A NxN Rotary Optical Switch Using the Right Angle Collimator
The right angle collimators are to be used in a NxN optical switch, shown schematically in FIG. 2, where light from a central fiber is directed to one of the several output fibers. The output fibers are aligned/directed out of the plane of the paper with their light reflected at right angles into the plane of the paper. The fibers may be in a bundle for instance. In the switch operation, the light from a central rotatable source is directed at one of several output fibers. The light from the central fiber sweeps in a plane as it is rotated and the output fiber entrance pupils must lie in this plane. The ferrule of the output fiber must be pointed at the input so that the center of the exit pupil of the rotatable source and the center of the entrance pupil of the output fiber must be aligned in 5 dimension, X, Y, Z, yaw and pitch. Assembly techniques can accomplish all but the yaw axis. Rotating the ferrule (or the ferrule/fiber) to the correct angle aligns the yaw axis. Using a power meter to maximize power transfer is a convenient assembly aid before the fixing in place is done.
The optical switches described in Fogs 1 and 2 deflect a collimated beam of light by reflecting it off a rotatable mirror. Rotation of the mirror causes the light direction to be changed by Snell's Law. The requirements of taking light from a fiber, collimating it switching it and have it enter another fiber requires a small, fast acting, rotary actuator capable of doing high precision angular positioning of a light beam to a few microns at a radius of 50 mm or so. The actuator preferably should be flat to make the switch as flat as possible though this is not an absolute requirement.
The Rotary Motor
The motors we will describe are built to operate on electromagnetic principles in which a current carrying conductor in a magnetic field experiences an orthogonal force i.e on the Lorentz Principle. This effect is used in voice coil motors used as head actuators in disc drives. This motor is shown schematically in FIG. 1. Here the coil, 14, on a bearing,15, operates in a magnetic field produced by two magnets of opposite polarity, 12, 13, shown here with vertical and horizontal shading. The slanted portion of each coil experiences a force orthogonal to the leg of the coil with a direction that depends on the direction of current through the coil. Using a pair of opposite polarity magnets ensures that the forces, and therefore the torques, add rotationally. The normal read head motor has a bearing that is off-set from the magnets but here the motor design can be improved in several dimensions. First the limited rotation can be improved by placing the bearing in the center of the magnet. This is an impossible position for a magnetic disc drive since the drive magnet would affect the disc coating but is acceptable in optical switches.The motor swings through an arc defined by the width of the coil. The generated forces will oppose if the coil is contained entirely in the field of a single magnet and net torque will drop to zero. This motor is un-commutated. FIG. 5 shows a coil/magnet arrangement that allows in the limit 180 degrees of rotation. FIG. 6 shows a design that has both extended rotation and double torque.
A normal, commonly purchased angular position transducer is used as a position pick off to give a signal that can be used in a normal servo feedback scheme to give the required positioning accuracy needed to feed the switched light through a lens into the second fiber; thus making a switch.
Typically voice-coil motors in miniature disc drives are designed to have a response time of about 1-2 milliseconds. These motor designs are capable of reaching a desired position starting from a first position in less than a millisecond. This is due to the multiple coils and the use of multiple magnets.
A Second Motor Design
The above motor is essentially planar and can be made very thin, of the order of 4 mm or so. Relaxing this constraint allows other motor designs. The first is to fold the coil over as shown in FIG. 7. Here the cylindrical structure,16 and 17 carries the coils. The magnets, 18, are polarized radially. The structure, 16 and 17, forms the rotor and is shown with a black top and gray body and is in the shape of a cup. The drive coils (in black) are on the outside of the cup. One coil is shown. Only the vertical legs of the coils generate force and therefore torque; the top and bottom parts of the coils generate no torque in the desired direction of rotation. Several coils may be used depending on the torque required and the range of angular movement required. FIG. 7 shows the input rotary mirror, 6, of FIG. 3, mounted on the rotor structure.
An NxN switch is made by marrying the ferrule of FIG. 4 to the motor shown in FIGS. 5, 6 or 7 as shown schematically in FIG. 8 to make a switch unit module, 20. The light path form one to the other is shown in the dashed line 19.
A full NxN switch is made by taking N ferrule/motor modules, 20, for the inputs and N ferrule/motor modules for the output and arranging them appropriately. FIG. 9 shows possible arrangements of the input and output modules. The arrangement on the left shows a rectangular base with inputs, 21, on the left and outputs 22 on the right. The black dots indicate placement of the directors. The longest, 23, and shortest, 24, paths are labeled. The longer the longest path, the smaller the variation in the path length, but the larger the potential optical loss. The circular arrangement has a shorter shortest path, 25, but a larger variation between the shortest and longest. Optical lossconsiderations make the circular arrangement preferable in general.
Dual Servo Control
The ferrule/motor module can be converted to a servo-controlled module by the addition of a position transducer and a controller. Both the transducer and controller are common, off the shelf parts.Dthe position transducer may be a digital optical shaft encoder available from a variety of sources or a position sensitive optical detector available from Hammamatsu. Both of these give a position with reference to some base position, and the controller may be a Field Programmable Gate Array (FPGA) or a microprocessor. The software in the controller uses position transducer 21 to servo-control motor 23 and position transducer 22 to servo-control motor 24. That is the controller is time shared between the servos. The controller is timeshared between the 2N motor/transducer pairs that make up the NxN switch. The serco software can be the common Proportional Integral Derivative (PID) system found in all elementary servo textbooks. The software running in the controller, i.e. the control system,
Drives the servo controlling a first motor, say 23, to point the optical beam to the second motor, 24 and vice versa. Doing this for all of the N pairs of motors makes the NxN switch.
Possible Errors I, Static Calibration and Adjustment
Anyone skilled in the art will recognize that assembly errors will exist in this scheme. The main errors that are possible are in positioning the ferrule to the motor so that the beam falls at the desired position on the rotating mirror, usually the center of the mirror, and in assembling the motor axis to be orthogonal to the base plane so that as the motor rotates, it sweeps the beam parallel to the plane.
These errors can be compensated for, as shown in two dimensions in FIG. 10 by using linear actuators, 28 and 29 placed between the motor mount and the base of the switch. Actuation of both actuators lifts and drops the motor while driving one up and the other down tilts the motor. Similar actuators in the other direction, say one in the x-axis and the others in the y-axis, plus the rotation around the vertical or Z-axis give allows roll. pitch and yaw three dimensional adjustments. They are adjusted.by miniature linear actuators that are conveniently implemented by piezo-electric devices. Other miniature linear actuator devices are also possible such as a rotary devices arranged to lift the motor plate by using a kinematic mechanism such as an eccentric cam or a linkage. Piezo-actuators are used in our preferred embodiment because their range of motion is small, their force is large and they can operate at high speed.
The errors described above can be compensated for at assembly time by directing the light to the second member of a pair measuring the intensity or received power and using conventional hill-climbing software to maximize received power. The hill-climbing software executes in the servo controller. One calibrated the measures values are stored for future operations and re-calibration.\
Possible Errors II, Dynamic Calibration and Adjustment
Compensation for dynamic errors can be achieved by the same principles if a suitable feedback transducer is employed. Industry practice teaches that 1% of the light in the beam may be used for such calibration purposes. FIG. 11 shows the entire light path of one port to another in the full embodiment of the NxN switch Here a pair of optical detectors, 31 and 32, are added, one to each ferrule/motor light director and attached to, ideally bonded to, and placed behind the rotatable mirror. The mirror is arranged to be 99% reflective allowing 1% to land on the detector. The preferred embodiment of the detector is a quad, or quadrant, detector ( Hammamatsu is a suggested supplier amongst several) that gives a signal corresponding to where a spot of light lands on its surface. This device tells if the impinging beam is centered or not and if not, the magnitude and angle of the deviation of the landing point from the optical center Any errors between the center of the mirror and the center of the quad detector can be measured and stored at calibration time for use in dynamic adjustment. Dynamic errors will appear as a dynamic offset of the quad detector. This signal is sent to the controller, 25, of FIG. 9. The control algorithm first positions the mirror by dead reckoning using its own position transducer. This gets light impinging on the detector of its target mirror. The signal from the target mirror is then used to position the first mirror using the linear actuators, 29 and 39 described in the discussion of FIG. 10. The same is true in reverse to position the second mirror, it uses first its own and then the target quad on the target ferrule/motor director. Again a PID control scheme may be used, with parameters adjusted for the sensitivity of the transducer, (sometimes called the transducer gain),
Conclusion
N multiple pairs of ferrule/motor directors positioned so that each input can direct its light to every output and vice versa, as on a circle makes the required NxN switch.
An Nx1 switch used one ferrule motor director and n statically aligned ferrules without motors arranged in either a circular, line or other suitable configuration:
Patent Application
20050041910
