Basics of optical incremental encoders - technical

Basics of Optical Incremental Encoders

An encoder provides information about the position of some movable part. For example, in a plotter the controller must know where the pen is at all times and issue appropriate commands to the motors to make the desired plots. This discussion will focus on rotary encoders, but the principles apply equally well to linear encoders.

HP produces transmissive optical incremental quadrature encoders. In these encoders, light passes through a patterned wheel and a detector senses the resulting shadows. An incremental encoder only reports relative motion. The controller must use other means to find its reference position, and then count pulses to keep track of the instantaneous position of the moving part.

A quadrature encoder has two output channels which together indicate both the distance and the direction of motion. The signal from each channel is a square wave, one cycle for each spoke of the slotted code wheel. Typical encoder output waveforms and performance definitions are shown in Fig. 1. Channel A leads channel B by 1/4 cycle for motion in one direction and lags by 1/4 cycle in the other direction.

There are four possible states of the two channels, occurring in the sequence ..., s1, s2, s3, s4, s1, ... in one direction and ..., s1, s4, s3, s2, s1, ... in the other. From any given state, only the two adjacent states are valid. An out-of-sequence state signals an error. Even in unidirectional systems, a quadrature encoder is often used for error checking and/or to double the resolution (four transitions per cycle instead of two).

A digital encoder gives no information about motion until a signal transition occurs. Digitizing discards some information from the analog optical signals. Some applications, such as stopping at a point (with no deadband), might benefit from using the missing analog information. Normally, however, the resolution (states per revolution) of the code wheel is made high enough so the uncertainty within a single state is acceptable.

Errors in the Output Signals

There is sometimes confusion about the terms accuracy, resolution, and repeatability. All three have something to do with knowing the position, but there are important differences. Accuracy is the relation between the reported position and the actual position. Resolution defines how small a motion can be detected. Repeatability concerns how closely the system can return again to a particular location. For example, imagine an encoder with a very high-count code wheel mounted on a bent shaft. Accuracy will not be very good, because the center of the wheel isn't where it should be. But the system can move and return to the same spot quite precisely (repeatability) or detect a small motion (resolution). On the other hand, a well-made encoder with a small number of counts per revolution would have good accuracy because its output pulses occur at the right places, but low resolution because the shaft has to turn a long way before the next transition.

Since the output signals are periodic, it is convenient to discuss their timing and errors relative to one cycle of those signals. Each cycle is defined as 360 electrical degrees. For a 500-count encoder, 360 electrical degrees equal 1/500 of a full turn of the shaft.

Cycle uniformity (position of corresponding transitions on adjacent cycles) is usually quite accurate, because the various sources of error affect corresponding transitions in the same way. A bent spoke in a metal code wheel does cause a cycle uniformity error, but even that error is decreased because the encoder averages the positions of several spokes. An eccentric code wheel will cause cycle uniformity errors that accumulate into position errors equal to the total eccentricity.

Ideally, the signals for the A and B channels differ in phase by 90 electrical degrees. Errors in this phase relationship can be caused by mechanical misalignment between the code wheel and the detector. In Fig. 2, the dark lines represent code wheel slots, and the cross hairs represent the detectors for channels A and B. Suppose the center of the code wheel is moved a distance e. The code wheel slots that should be over the detectors are no longer in the correct positions. Rotation of the code wheel can still align the slots with the detectors, but it must rotate counterclockwise for A and clockwise for B. The difference between those two positions is the phase error between the A and B channel signals. Reference 3 listed on page 106 calculates this error to be [delta][phi] = (360/2[pi])eSn/r.sup.2., where [delta][phi] is the phase error in electrical degrees, e is the misalignment, S is the separation between the A and B detectors (near zero in the HEDS-9000), n is the cycles per revolution, and r is the radius of the code wheel.

Errors in pulse width are typically caused by threshold or balance errors in the detectors, or by a nonuniform light source. Pulse width and phase errors combine to cause errors additively in at least one of the four states.

If the code wheel happens to stop right at the point of a transition of the output signals, the output should not chatter or dither between the two adjacent states. Such dithering should not in theory cause the controller to lose count, but it would be a burden. The remedy is to put hysteresis in the logic circuit. This requires the code wheel to travel slightly beyond the ideal switching point, then to switch abruptly. To return to the previous state, the wheel has to go back across the transition point a finite distance before the output will snap to that other state. The amount of hysteresis should be large enough to avoid dithering yet small enough not to add significantly to position uncertainty. In the HEDS-5000 and HEDS-9000 encoders, a built-in circuit provides the required hysteresis. but is a useful in-process test point for monitoring the optical system.

Optical Balance in the Emitter/Lens System

Differential detection accommodates large variations in light level, but requires matching between the two halves of the system. Imbalance in the optical path or circuitry creates pulse width errors. Although it might have been possible to adjust pulse width at final test, unit-by-unit adjustments clutter the path to manufacturing simplicity. The design stategy for the HEDS-9000 was to make the optics and detector inherently balanced. This was not particularly difficult in the detector, but in the emitter system this was technically challenging.

The detector design takes advantage of matching components that comes naturally in an IC.IC resistors, for example, may have 20% variation from batch to batch, and a significant temperature drift, but adjacent resistors on the same IC usually match each other within 1%. The circuit leans heavily on characteristics that come free and avoids unrealistic expectations on process control.

One figure of merit for encoders is their tolerance for uncertainty in the distance from the code wheel to the detector. Such tolerance allows the system designer to use cheaper code wheels (less flat), cheaper bearings (more end play), or less careful assembly. The code wheel should not rub on the detector at its closest position, yet must cast a clear shadow at its farthest position. This leads to another restriction on the light source: it must produce a collimated beam of parallel light rays. The HEDS-9000 emitter lens is a precision-molded polycarbonate lens with an f-number of 0.7 for high efficiency and two aspheric surfaces for control of collimation and uniformity. The emitter, a light-emitting diode (LED), produces more intense light on-axis than at wide angles. The lens funnels more of the weaker wide-angle light to each unit area of exit surface. It does this compensation while causing more than 50% of the total light from the emitter to emerge in a parallel, collimated beam.

Tight collimation requires a small light source as well as precise lens surfaces. The gallium arsenide phosphide emitter radiates from a 60-[mu]m-wide active area on its top surface.

Emitter system characteristics are monitored in the manufacturing process to ensure consistent collimation and unformity. Fig. 3 shows the intensity profile across the diameter of the light beam. The dip in the middle of the graph is the shadow of a test wire at a distance of about 45 wire diameters. The contrast between the shadowed and nonshadowed areas remains good.

Package Design

The package design is illustrated in Fig. 4. The package carries the IC, the LED, the lens, and a current limiting resistor. It provides electrical interconnections between these elements and includes connecting pins to the outside world. The package also aligns the emitter system to the detector and provides mounting location features. A plastic body molded around a stamped metal leadframe was developed for the HEDS-9000 because it performs all of these functions, replacing as many as 10 separate parts (see Fig. 5), saving labor, and improving reliability.

The molded leadframe is flat for die attachment and wire bonding of the detector IC, the emitter, and the current-limiting resistor. The detector is given a protective transparent encapsulant. The collimating lens is inserted over the emitter cavity and sealed. The package is then folded into its final C shape and tested. Five 0.025-inch-square posts on the leadframe become the connecting pins. Details on the molded plastic align the emitter system to the detector. Holes provide for screw mounting and registration of the module by the customer.

Manufacturing Technology

The assembly process for the HEDS-9000 is similar to that used for a standard IC package. Since IC manufacturing is a well-developed technology, off-the-shelf machines are widely available for high-volume production.

The LED, the IC, and a resistor die are automatically positioned and wire-bonded using standard equipment. IC encapsulation, lens insertion and sealing, package folding, and testing operations are done on equipment specifically developed for the HEDS-9000.

Because the HEDS-9000 is not dependent on special alignment by the customer, it can be 100% tested in an as-used configuration for both electrical characteristics and encoding parameters. Testing is performed at over 700 units per hour with a tube-to-tube parts handler. To test a 1000-cycle-per-revolution encoder at the required resolution of 1/360 of a cycle, the tester must resolve about 4 seconds of arc (0.000017 radians). A code wheel turning at constant speed allows the tester to make time measurements rather than angle measurements. Each output channel is monitored at a 50-MHz sample rate.

Production System

Work in progress (WIP) flows through the HEDS-9000 production line under a pull system, that is, WIP is processed only as required to satisfy the needs of the next operation. Rework, hot lots, and fill jobs are not allowed. The result is short lead times without huge inventories. 1,000-piece orders are delivered in less than four weeks. Pull systems and JIT (just-in-time) manufacturing are common throughout HP and other manufacturers, but there are some difficulties in applying a JIT philosophy to component manufacturing. Because of special process steps, such as oven cures, components are best made in batches to allow better use of equipment and labor. HEDS-9000 production machines are loaded using magazines holding 48 units each. This accommodates batch processing while allowing the benefits of a pull system.

Fig. 6 is a simplified diagram of the HEDS-9000 production line with workstations and WIP holding areas. A workstation is an operation or combination of operations under the control of one operator. The workstations are balanced, that is, all stations run at approximately the same speed. WIP flows from a cabinet through a workstation and into the next cabinet on a first-in, first-out basis. WIP cabinet contents are not allowed to exceed a preset maximum. Limits are set to allow a certain amount of a preset maximum. Limits are set to allow a certain amount of random speed variation and machine setup time without stopping the line. For example, if station 3 shuts down long enough, cabinet B fills to its maximum and station 2 cannot work. This propagates backward, so one person's problem becomes everyonehs problem. A cabinet that is always full indicates that the next station is a bottleneck and it receives engineering attention. If the speed at a bottleneck is increased by 50%, the output of the entire line increases by 5%.

The pull system works best if WIP has few reasons to stop before completion. No rework of defective units is allowed. Effort is spent learning to make things right rather than learning to rework. Yield is high. Quality problems are painful and get fixed in a hurry. Cycle time has been reduced by eliminating time-consuming steps. For example, conventional adhesives and coatings would require a total of about 14 hours of oven cure time. To eliminate one oven cure, an adhesive was developed that cures in less than 5 seconds when exposed to ultraviolet radiation. Cumulative cure time was reduced to 2 hours.

Reliability

Table I summarizes the reliability test results for the HEDS-9000. Reliability is inherently good because of the low number of parts and interconnections. Here again, the benefits of integrated design are reaped. Judicious choice of materials, stress-test monitoring of production parts, and statistical quality control result in consistent performance.

Encoder Selection

There is no one best encoder for all applications. Assuming that an optical digital incremental encoder like the HEDS-9000 is appropriate for the application, some remaining important choices involve code wheel diameters and counts.

Larger code wheel diameters give better accuracy, but take up more room and have greater inertia.

Higher code wheel counts give better resolution, but at a higher signal frequency, which could approach the limits of the detector or the controller. Count and diameter combine to determine feature size. Cost can be largely dependent on area, a fact that favors small diameters until a point is reached where the feature size gets too small. Code wheels with very small features are hard to make, therefore expensive.

All encoders in the HEDS-9000 family share the same temperature and frequency capabilities. Different versions are available for various code wheel sizes and resolutions. The HEDS-9100 module is used with 28-mm-diameter code wheels. The HEDS-9200 is for linear applications using a code strip rather than a rotating wheel. The part number HEDS-9000 specifically refers to modules used with 56-mm-diameter code wheels but is also used as a family part number. Many standard resolutions are available and specials between 1.2 and 8 lines per millimeter can be provided by tooling a new photodiode layout on the IC.

Acknowledgments

Mark Bullock conceived much of the manufacturing architecture. Chris Togami designed the leadframe and plastic parts. Joe Dody and Donald Lapray contributed in the areas of processes and tooling. Art Wilson did prototype tooling. Thomas Lugaresi provided IC layout direction. Akhtar Khan's analysis provided the insight that led to the detector IC's ability to operate smoothy over its wide temperature range. Thomas and Akhtar brought the insight to fruition with simulation and physical modeling. Craig Sue, whose leadership was essential for keeping the whole IC task on schedule, was responsible for IC product engineering, including test development and IC characterization. Curt Wilson did product characterization and ensured the integrity of the data sheet. Richard Ruh managed the project in its development phase with unrelenting dedication. He took over from the able leadership of Debbie Haferburns, who left to manage her family, also in development. Larry McColloch gets credit for the pull sy stem and much ingenious tooling design. Victor Loewen and Bill Bilobran supplied expertise in many areas including tester design. Dave Oshima, Ray Tam, Yoshi Tatsumi, and Maria Costa made successful production possible. Bill Loesch and Bill Beecher provided unwavering support and sage advice from the inception of this project to product release. Bob Steward brought in the right people to make it all happen. Lui Kok Chwee, George Lim, and many others in Singapore established high-volume production capabilities. Karen Owyeung and her predecessor Lisa Wade made certain that the customer viewpoint predominated.

COPYRIGHT 1988 Hewlett Packard Company
COPYRIGHT 2004 Gale Group