New techniques align fibers

On-the-fly measurements and accurate mechanical control ensure the proper alignment of optical fibers.

Joseph Ting, National Instruments, Austin, TX -- Test & Measurement World, 5/1/2001

The booming market for fiber-optic communication systems requires the rapid production of optical components in large quantities. In many cases, producing these components involves accurately mating an optical fiber to another fiber or to a device such as a waveguide, filter, or transmitter. The demand to rapidly produce components means fiber-optic (FO) equipment manufacturers can no longer afford to rely on relatively slow “low-yield” human workers. Instead, these manufacturers have started to use equipment that combines automated production and measurement equipment in a complete system.

Figure 1. A prototype fiber-alignment system uses feedback from optical instruments or cameras to position one optical fiber for maximum light transfer to the other fiber.

To study the challenges created by combining many technologies in this type of production equipment, a team of engineers at National Instruments built a prototype optical-alignment system (Figure 1 ). Our team used standard off-the-shelf hardware and software, and you can adapt our techniques to many applications that require interactions between measurement and mechanical equipment.

Whether you aim to mate two fiber ends, or a fiber and a device, the techniques involved are generally the same. You want to position a “transmitter” and a “receiver” to ensure that as much light as possible travels from one to the other. (In our prototype, we only considered mating two fibers.) During alignment, one fiber remains in a fixed position while an electromechanical stage moves the other fiber in small steps.

Our system uses active alignment, where one fiber connects to a photo detector and the other fiber acts as a light source. In short, a PC moves the stage along the x, y, and z axes to maximize the light transferred from one fiber to the other. After properly aligning the fibers, another system—not included in our prototype—can bond the fibers to produce a minimum-attenuation link.

Run a coarse alignment first

Active alignment uses a coarse scan followed by a fine scan. The coarse scan moves one fiber until it detects the general location of the narrow beam of light emitted by the other optical fiber. A fine scan then pinpoints the maximum coupling of light between the two fibers. An alignment system also can perform the coarse scan and position the fiber on the stage using cameras and machine-vision algorithms in preparation for the fine scan. A machine-vision system can also perform a coarse scan and it is often simpler and faster, but you can apply it only in a few cases. I’ll explore both techniques.

Our team started designing an automated-alignment system by choosing the best mechanical hardware. Current nano-positioning actuators include servo motors, stepper motors, linear motors, and piezoelectric devices. Each actuator type has an associated precision range, travel range, maximum velocity, torque, and service life. And keep in mind that mechanical variations in lead screws, gears, cross-roller bearings, and air bearings can affect operation. The specification and tradeoffs of mechanical components go beyond the scope of this article, but equipment suppliers and mechanical engineers can offer you guidance.

We wanted to properly align a single-mode fiber with a core diameter of less than 10 µm, so we chose a stage that could position a fiber with a 50-nm accuracy. The stage we chose could travel 25 mm in the x, y, and z axes at up to 6 mm/s. (We required the stage to carry a 1-kg load and to operate continuously for a year.) We chose a servo-based positioner that includes Renishaw (Wotton-under-Edge, UK) linear encoders that provide position feedback to our motion-controller board. But because mechanisms aren’t perfect, the motion controller needed position data to ensure it knew the real position of each actuator.

When it comes time to start the alignment process, we first use the stage to coarsely align the two fibers. The coarse alignment aims to detect “first light” in an area that centers on the stationary fiber, from a few hundred microns to a few millimeters across. Most of today’s automated systems move a fiber to a preset position in the search area, stop the motion, and make an optical-power measurement. The system repeats this “move-stop-measure” process until it detects light.

Home in on the light source

Figure 2. (a) The back-and-forth path and (b) the rectangular-spiral path require motion on only one axis at a time. Thus, almost all motion controllers can drive a positioning stage using these patterns. More complicated paths often require simultaneous motion in two axes.
Figure 3. The light intensity at the end of a fiber follows a 3-D Gaussian curve.
Figure 4. Low-discrepancy scanning uses an algorithm to reduce the average coarse-search time. Although the path looks random, it follows a definite, calculated path.

Systems typically use back-and-forth or rectangular-spiral search paths (Figure 2 ). The rectangular spiral is theoretically superior to the back-and-forth search because although both paths have the same average search time, the rectangular spiral produces less variance (standard deviation). In practice, the performance of a rectangular spiral may suffer slightly due to inertial-settling caused by frequent direction changes.

Typically, light emitted from a single-mode fiber produces a beam with a 3-D Gaussian profile (Figure 3), and inaccuracies from positioning the stage at the start of a coarse scan generally fit a circular probability distribution. The circular nature of these measurements means a circular spiral may offer a better coarse-alignment path. Currently, automated systems that produce this type of circular scan approximate a spiral by connecting straight line segments. This approximation introduces some error, but it places measuring points near enough to the spiral path. An Archimedes (radius = aQ) spiral proved a good choice for us to use for evenly distributed sampling, but you can choose from other types of spiral paths, too. A logarithmic spiral (radius = e^aQ), for example, produces a scan that increases the density of points as it approaches the center.

Existing scanning methods work adequately for coarse alignment, but better algorithms such as low-discrepancy scanning (Figure 4), based on the Halton sequence, can further decrease the average search time, by a factor of two or three. The move-stop-measure, or discrete measurement, process causes the greatest bottleneck in automated systems. To overcome this bottleneck, we used a synchronized and continuous process that increased the performance of the prototype system.

Identify the bottlenecks

First-generation automated-alignment systems are limited to a discrete-measurement process for many reasons. First, the systems use stand-alone motion controllers and instruments that communicate through serial or IEEE 488 buses. Those buses, and even the newer HS 488 bus, can’t keep up with real-time control needs.

Second, stand-alone equipment often lacks a means to deterministically trigger measurements. Thus, the controlling PC must halt the stage, send measurement commands to the instruments, and wait for a response before moving to a new position. These steps take time. The move-stop-measure technique cannot guarantee true position measurements because servo motors and actuators can jitter while settling to a stop, and stepper motors can slip.

Finally, first-generation systems cannot move continuously along arbitrarily chosen paths such as a logarithmic spiral. Only a few high-end motion controllers provide the necessary “contouring” operations that spline through the coordinates of arbitrary curves.

To overcome the limits of older equipment, we based the electronic portion of our prototype on the PCI bus and the superset PXI bus. The PXI bus includes the real-time system integration (RTSI) bus that synchronizes operations among PXI cards. We used a PXI-7344 four-axis motion-controller card for contouring and high-speed position registration. We chose the PXI-6052E, a high-resolution 16-bit data-acquisition board, to acquire hardware-timed power measurements from a DLPCA-200 transimpedance amplifier from Femto (Berlin, Germany). The amplifier connects to an InGaAs Model DET410 photodetector from Thorlabs (Newton, NJ).

Base readings on position

To synchronize actuators and measurements, we used a time-based measurement instead of position-based measurement. Thus, measurements occurred at specific times rather than at specific places. Most position-based measurement systems rely on position encoders or position breakpoints to start measurements. But encoder and breakpoint triggers work on only one axis at a time, so we decided not to use time-based measurements.

We used a hardware clock signal on the RTSI bus to trigger measurements. The system acquires optical-power measurements at a set rate and simultaneously collects position data from the sensors on the stage. In this setup, we obtained true position data that accounted for any errors due to inertia or slippage. This triggering technique lets our computer make about 80 synchronized measurements/s, which is from 8 to 40 times faster than if we used a serial or IEEE 4888 instrument. A coarse alignment that took 15 min now takes less than 1 min.

After our system performs a coarse alignment, fine-alignment steps begin. The fine alignment uses a dynamic “hill climb” algorithm that works much like the Newton-Rhapson method of iterative approximation. In effect, the software moves the stage along a single axis and acquires optical-power until power measurements decrease. Then, the software halves the step size and scans in the reverse direction until power decreases on the other side of the peak. The process goes back and forth in smaller and smaller steps until it stops at the optical-power peak.

The hill-climb algorithm aligns only one axis at a time and is sensitive to certain types of noise, but it offers a reliable high-resolution result. The PCI-bus hardware in our system uses the hill-climb technique and can run a fine alignment in under 10 s. The same alignment took 60–90 s using IEEE 488 instruments. (New algorithms may reduce fine-alignment times.)

Incorporate machine vision

Machine vision can also play a role in fiber-alignment applications. A camera can capture thousands of pixels from which software can extract accurate positioning data. As a result, a machine-vision system can perform a coarse alignment faster than alignment systems based on optical-power measurements. And visual measurements can eliminate the possibility of collisions between fibers, something measurements based on optical power can’t ensure. Unfortunately, only a few configurations of fiber-alignment equipment provide the right type of conditions for a machine-vision system. Those arrangements provide unobstructed views for cameras and operators, proper lighting for cameras, and enough space to mount the cameras.

We calculated that we needed a resolution of about 200 µm/pixel in a 640x480 image to properly locate the ends and positions of the fiber. At this resolution, we could ensure a coarse alignment that would couple 10% or more of the light from the transmitting fiber to the receiving fiber. Because a single camera can make only 2-D measurements, we used two cameras placed 90° apart, parallel to the axes of motion, and perpendicular to the optical axis.

Figure 5. For a coarse alignment, machine-vision algorithms can locate the tips of optical fibers and calculate their coordinates quickly.

The vision-based coarse alignment is straightforward to prototype and develop using machine-vision software tools. A few high-level commands locate the 2-D pixel coordinate of both fiber tips. By using a top view ( Figure 5) and a side view, the software can calculate 3-D coordinates. Usually, a single set of images provides enough information for the system to bring the fibers close enough to couple light. At that point, fine alignment based on optical power can take over.

Selecting timed measurements that simultaneously gather optical and position data helped us produce a prototype that would properly align optical fibers faster than existing systems. The combination of accurate electronic measurements and motion control may apply to other inspection and production tasks. T&MW

Joseph Ting works as a motion-control product-support engineer at National Instruments. Prior to joining NI, he built test systems at VI Technology. He has a BSEE from the University of Texas at Austin.