Optical Encoder Project

Many years ago, I competed in MASLab, an MIT competition to build an autonomous robot. A year or two after that, I was on the MASLab staff, helping prepare the hardware for the year's competition. One of the projects I worked on was an optical encoder. This optical encoder was somewhat unique in that it was designed to be very robust (or at least substantially better than the disaster that most hobbyist designs are). It worked under a wide range of lighting conditions. The disk could be anywhere from a couple of millimeters to several inches from the optical elements. The pattern on the disk could be printed on a wide range of printers. It was also designed to be very inexpensive, and easy to build (few parts to solder). It was not designed to have particularly high resolution. To achieve these goals, I used a differential design, where the encoder looked at two patterns, and compared which was brighter, rather than just looking at a single pattern.

In late 2005, I received an e-mail from a student at another university asking for my design. The student mentioned that there weren't much information on the Internet about creating optical encoders, so I figured having my ready-made design up might be useful, and as a result, I made this page.

Optical Encoder Overview

If you are familiar with optical encoders, skip this section.

An optical encoder measures how far something has moved, optically. It does this by looking at a pattern of alternating black-and-white, and counting number of transitions.

Optical encoders come in two varieties: regular and quadrature phase. A regular encoder uses a simple pattern of black-white. It can count number of transitions, but cannot tell the direction of motion:


PostScript source

In contrast, quadrature phase uses two staggered patterns, so that the system can tell which way the thing is turning. This can either be done explicitly with two patterns, or implicitly by using one pattern, but appropriately spaced sensors. An example of a two-pattern disk:


PostScript source

In addition, optical encoders can use one of two types of sensors: reflective and break-beam. Break-beam tend to be more accurate, but require precise mechanical alignment, sometimes hard to achieve in hobbyist projects:


Break-beam sensor. Image from Robot Store (used without permission)

In contrast, a reflective sensor just consists of an LED next to a matched photosensor (photodiode or phototransistor).

Quick Guide

The circuit used was:

xcircuit source
An improved circuit would have added Schmidtt triggering to guarantee smooth transitions, as well as a low-pass filter on the output.

xcircuit source

I chose the simpler design primarily in order to reduce the amount of soldering students had to do.

One more step up, the resistive divider would be replaced with a potentiometer, so that the user could tune for mismatch between photosensors.

In addition, I have a fairly flexible PostScript file that can generate disks for optical encoders. It is easy to modify it for different sizes, regular, quad-phase, single-ended, differential, and other designs. If I recall correctly, I based it off of someone else's optical encoder PostScript file (for generating simple single-phase disks), but it was adequately long ago that I no longer recall whose:


PostScript source

The parts we used aren't very important, but for reference:

We made a PCB out of this. I had an old version, where some of the holes for the parts were not big enough. I believe the current version is this: tar/gz-compressed ZIP-compressed, but I would verify it prior to use. The pcb was designed using the free PCB package. This was my first PCB design. If I recall correctly, my fab didn't handle circles properly in the Gerber file, so I needed to modify the source of PCB to output approximations of circles made out of lines.

The alignment of the disk with the sensors is critical.

Details

A typical optical encoder will either use a break-beam sensor, or a single pattern (dual for quad-phase) of alternating black and white. This is done because most designs are in a well-controlled environment, where there are reasonably well known, fixed levels for dark and light. In this case, we had a far less controlled environment:

For these reasons, traditional hobbyist optical encoders tend to be complete disasters. To mitigate some of these problems, we used a differential design. In so much as the photosensors matched, so that the design was very insensitive to ambient light level, or how much light was reflected from the surface, so long as different amounts of light were reflected.

Using the phototransistors back-to-back as a voltage divider (rather than amplifying first) let the design to have a huge dynamic range.

The major disadvantage of the differential design was decreased resolution. We needed four rings for a quad-phase design, instead of two. As a result, the inner-most ring was on a much smaller radius, and so could not have very dense stripes. In addition, the photosensors needed to be reasonably well matched.

Traditional Photosensor Design

If you are planning to build a single-ended encoder:

You can buy packages containing a matched transmitter and receiver. These are usually only available in SMT, but are much more convenient. They come in both reflective and break-beam versions. Some are just the transmitter/receiver pair, but some include a complete circuit that can, assuming some minimum rate of rotation, self-tune itself.

The traditional photosensor design is:


xcircuit source

A slightly better design, if you know some minimum rate of rotation, is:


xcircuit source
The inductor can be either a true inductor, or an active inductor-like device made from transistors. This removes the DC (ambient) light level. The important parameters for the op-amp are:

This would be typically be followed up by a second stage consisting of a low-pass filter and a Schmidtt-trigger to clean up the signal.

There is one more simple way to get an optical encoder for hobbyist projects -- use a computer mouse. Plenty of household electronic devices include optical encoder (my old, broken CD-changer had at least 3 or 4), but mice are:

The disadvantage is that, in contrast to a normal optical encoder, some PS/2 mice may integrate error over long periods of time. As a result, while they work very well for sensing speed, they do not work as well for sensing position in a long-running application.

Performance

I was not around during the competition, so I do not know how well they performed in the field, or what the problems, if any, were. I initially heard from the contest organizers that the design was working very well. Later, a former student told me that some number of students having significant problems with getting reliable readings. I do not know how common problems were, but I would suspect the dominant problems would have been related to:

The version I had in my lab was extremely reliable. Without bypass capacitors, it could still tolerate a 1V sine wave on the power rail at an arbitrary frequency. It could work if the disk was as close to the sensors as possible without touching, as well as half a foot from the sensors. It could work in the dark, as well as with a bright light shining directly at the disk.