I recently installed a closet door ball catch mechanism from Prime-Line Products on a folding door. The device is there to help make the door stay open unless I want it to close, and vice versa. You've probably seen these devices in your favorite home improvement store — or perhaps your own house. They are fairly common.
Installation was straightforward, and I especially appreciated the unit's very simple yet effective 3/8-inch (1cm) trim adjustment, which allows you to accommodate the gap between the door edge and the track and decide how easy or firm you want the detent mechanism to be.
It's nice that these mechanical devices have such trims, but it is not just mechanical designs which need them, of course.
Compensation for offsets, nonlinearities, and other errors plays a big part in analog-signal channel design, especially when there are sensors involved. How you implement this calibration can become a major system design issue, since there are many legitimate (and some inadequate) ways to do so. Methods range from having a PC board with too many trimpots to having a board with all the circuitry on one IC and no visible means of adjustment.
Back in the day, before microcontrollers ruled the land and before memory was available, cheap, and integrated, the focus was on perfecting the analog circuit. It was serious work to look at all the sources of error from the sensor onward. It involved studying data sheets and working out an error budget. You had to look at typical specs (briefly) and then really dig into worst-case specs at nominal temperature and operating conditions. This meant you would consider the tolerance range of passive components and the minimum and maximum values of the parameters of active components. Then you had to go further and factor in temperature drift and even aging if you were really serious.
With all the numbers in front of you, you could work out the worst-case error by using an root sum of squares (RSS) analysis or by summing the individual worst-case numbers. The RSS method is more probability based. It's not likely that all parts in a specific circuit will have specs that are at the extremes of the range of possible values. But for the situations where you think that might actually happen, you add up the individual worst-case numbers.
This analysis could be pretty tedious and painful, but a well-designed circuit, with careful attention to detail, could work amazingly well. Of course, sometimes designers used other strategies besides good, better, and best components. In some situations, a clever topology could make the unavoidable errors cancel each other out using components with opposing temperature coefficients (tempcos) or matched components with identical tempcos in a differential circuit. If you can't beat them, join them.
In part 2, we'll look at an engineer who was a genius dealing with component variations and doing designs that, once built and initially calibrated, needed little or no attention. And we will take a look at calibration strategies for devices with all their functionality built into one IC.