A few years ago, an instructor from the Fiber Optics Association complained to me about an ad from a fiber optic instrument company. He was indignant that it said, “Our OTDR Is Smarter Than You Are!”

I understood his indignation; I felt it too. Between the two of us, we’ve written dozens of articles and a couple of books about optical time domain reflectometers. We had even done some joint research into OTDR measurements compared to insertion loss measured with a light source and power meter.

Perhaps our indignation was justified then, because OTDR software purporting to be “smarter” than the typical user was often fooled by some of the quirks of OTDR trace analysis. Today, that claim may be justified for many modern OTDRs due to the microprocessors and software improved by years of advancements in artificial intelligence (A.I.).

It reminds me of the development of automobile transmissions. I learned to drive on manual transmissions, first “three on the column” in American sedans and pickups, then “four on the floor” in the foreign cars my brother sold. It was a sense of pride to be able to deftly manage clutch, gas and shift lever to produce a nice smooth change; even more so on downshifts where you had to “heel and toe,” nudging the gas to match engine RPMs for the lower gear.

But after a few years of living in the city, shifting became more of a chore. Driving involved hundreds of shifts and weariness in the left leg from so much clutch action. Automatic transmissions made more sense, even if they were inefficient and sluggish.

Automatic transmissions were transformed by microprocessors and software as greater efficiency was needed to conserve gas and improve performance, especially in smaller cars. These changes even made automatics the choice for racing and sports cars. Try buying a high-performance car with a manual transmission today; you will find your choice quite limited. Even the new midengine Corvette only offers an automatic transmission!

Software has gotten much more sophisticated, and A.I. software can learn from input data. Give it enough data, for example OTDR traces and information about the fiber being tested, and algorithms can learn the correct way to analyze the traces for that fiber. Eventually, you should have software that can correctly analyze the OTDR trace with high accuracy, comparable to a knowledgeable human operator.

But it has taken awhile to get there. Early OTDR “autotest” functions were notorious for producing bad results. I remember one user sent us a set of traces given as documentation of an installation’s completion. The data showed every fiber passed the autotest, while the traces showed the fibers were too short for the OTDR to even distinguish them. Another instance had the operator convinced that all the fibers in a cable were bad, when the problem was actually bad splices.

Those kinds of results led us to warn techs to not trust the typical OTDR autotest function. If you wanted to ensure valid results, we warned, a qualified tech who knew the instrument well should verify the autotest function was giving valid data on a couple of fibers before using it on other fibers in a cable.

But technology, especially A.I., forges ahead. A.I. software is powerful enough to learn to work like a tech does, which makes an OTDR smarter. Earlier OTDR autotest software chose “average” parameters in the OTDR setup to make tests, and as statisticians like to say, nothing is really average, so the setup was rarely ideal.

A.I.-based OTDR software does several tests on a fiber, varying range, pulse width and other parameters to choose the appropriate setup. It can even use shorter pulses for the fiber near the OTDR and longer ones for the same fiber further away, and use algorithms to integrate the data. Multiwavelength testing can distinguish bending losses from high splice losses at a splice closure. Using reflectance, it can determine whether a joint in the fiber is a splice or a connector.

These A.I.-based OTDRs are programmed to act like a well-trained tech and make decisions based on testing results. That’s something to be happy about, because our experience in training techs on OTDRs has shown that learning these complicated test instruments can be challenging for even the best techs.