[Date Prev][Date Next] [Chronological] [Thread] [Top]

Analog vs Digital Technology



  I've just read information from ES&S comparing the ES2000 (meaning our beloved Accu-Vote OS) to their new "digital" M100 poll level optical scan tabulator.  They are inaccurately referring to our technology as being analog and suggesting that this is the reason that we don't accurately pick up check and X marks.  First, some background on the terms analog and digital.

  Analog refers to a signal that can be of any strength relative to a scale.  Think of the old "analog" meters where a needle swings according to a signal strength.  Most things that can be measured start out as analog information such as time, speed, brightness, loudness, frequency, and so on.

  Digital refers to information that can be expressed as a discrete number.  This means that this information can be handled the same as any other number using digital processing technology.  Most computers and all of our personal computers are "digital" in that they directly handle these numbers.  And as soon as we "read" something analog, we have digitized it.

  That said, every processor at some level must use analog technology.  This is because the voltage or frequency carried by a wire or a fiber optic cable or a telephone line or whatever is analog by nature.  The difference between analog and digital technology is that analog must maintain the relative strength of a signal because it's the signal strength that contains the information.  With digital technology, the signal strength can be anywhere within its discrete step without introducing errors known as noise.  In fact, most digital signals are interpreted to be either on or off and the circuitry is designed to prevent the signals from being in the no-man's-land between these 2 levels.  Confused?  A picture would help but for those familiar with a sine wave, this is a typical analog signal.  The typical digital signal is the square wave.

  If you've followed me so far, we can address the question as to whether the ballot scanning machines, both ours and theirs, are analog or digital.  I hope that you can see that they must be both and that the only differences would be in when the analog light signal is converted into a digital signal.  For this, we must understand the different sensor technologies used to measure the amount of light reflected off the ballot paper.

  Our sensors are based on solid state semiconductor light sensors (specifically photo transistors) that convert the brightness of the light that reaches the sensor to an analog electrical voltage.  In the Lucid visible light readers, this voltage is then measured at regular intervals and converted to a digital number.  In the older infrared readers, additional analog signal processing was performed that resulted in a digital value indicating whether the sensor detected a mark or not.  Thus the visible light readers are more digital than the infrared readers.

  The M100 readers use charge coupled device (CCD) sensor technology which is essentially the same that that used in modern television cameras, photo scanners, and fax machines.  CCD devices work by accumulating an electrical charge at a rate proportional to the amount of light shining on them.  Thus to use these sensors you start by clearing the charge, exposing them to light for a specific amount of time, and then measuring the accumulated charge.  The amount of charge is an analog value that is digitized much the same as the electrical value produced by our photo transistors.

  Thus their system is no more "digital" than ours.

  So what's all this about reading check and X marks?  For this we have to consider resolution and interpretation.  Resolution is a measure of minimum size of things that can be "resolved".  Consider your last visit to an optometrist that had you read letters of decreasing size from a chart a fixed distance away until you couldn't accurately distinguish the letters.  This was a measure of your eye's ability to resolve the component shapes of the letters.  When it comes to resolving marks on a ballot, most of us have extremely high resolution and we can identify all kinds of different marks and this helps to make us good at interpreting those marks according to the voter's intent.  Building a machine that comes close to this ability remains well beyond our current technology.  For example, consider a voter that circled the candidate's names.

  What makes it difficult for machines is the extreme complexity of the rules and knowledge that we humans apply when interpreting the marks.  This type of "visual processing" is being actively researched in the area of "image recognition" in artificial intelligence labs around the world.  In the present we as engineers have to devise ways of simplifying the problem down to something that machines can perform accurately at reasonable cost.

  In our system, we simplified the task by dividing the ballot into quarter inch square cells and then concentrating on the center of those cells.  It's rather like the way that our eyes have much better resolution in the center of our focus than around the periphery.  We use the timing marks down the sides of the ballots to identify the centers of those cells and then we measure how dark it is in a small area around that center.  The more of this area that is covered, the darker the area appears.  We set a minimum darkness value for which we are assured of a mark.  If there is a light mark there, we may be able to report it as an invalid mark.  With this design, we can use a variety of ways of indicating the voting position or target area to the voter including ovals, squares, and even the ES&S style broken arrows.  Most any line through the center of the target area will be detected as a valid mark.  Check marks are difficult because they tend to go around the center.  X's are likely to be picked up if they go through the center.

  I don't know the internal details of how the ES&S system works but based on their technology and claims, I can speculate.  They probably have a higher resolution in that they "see" spots rather smaller than ours.  How small depends on their optics and the specific CCD sensors used.  Other than using a smaller spot, their system works much the same as ours.  Working with smaller spots means that you work with more spots and therefore you need more processing power.  The ability to reliably detect check and X marks then depends on keeping these marks in an area that is free of other marks that would cause false reads.  They suggest that they are doing this with their broken arrows.

  The ability to reliably detect checks and X's is generally a tradeoff against picking up false marks which come in the form of smudges, folds, dirt, dust, surrounding borders and text, and inadvertent marks.  We could modify our ballots and our software to better pick up these different marks but only at the expense of a significant change in the way that we indicate the target area to the voter.  We find that voters generally prefer our ovals over their broken arrows and so it becomes mostly an issue of voter education.  If the customer's voters are already trained to use the broken arrows, they probably have an advantage.  Otherwise, I expect that the advantage is ours.  In Saturday's election, out of some 4000 ballots, I had about 10 ballots rejected as overvoted because the voter had put the tiniest dot possible in the center of an oval while they were thinking and they were quite surprised that we picked it up.  I also had a few ballots returned with invalid marks because they had used their ball point pen to make check marks.  No ballots were returned as blank unless they were truly unvoted.

  So the issue of handling checks and X's is mostly one of ballot style rather than scanner technology.  You can try using a ball-point pen and different types of marks to see what I mean.  If your mark goes through the center, it'll likely be counted.  If not, it'll probably be returned as invalid.  If we don't detect it at all, it's probably too close to the oval edge.

  One last point, the infrared readers were very sensitive to the orientation of the ballot.  This was due to the way that the analog circuitry allowed for variations in the reflectance of the background paper.  The Lucid readers are no more sensitive to orientation than the ES&S readers would be.

  And ES&S accuses us of "telling half truths" about our system vs. theirs...

              Guy