原创 Circuits with interactive adjustments

A bicycle's front-and-rear derailleurs are an amazing mix of effectiveness, light weight, reliability, and tightly focused function, but they come with a frustrating initialisation phase. I was reminded of this a few days ago, when the cable that shifts the front derailleur on my bike snapped from wear. Fortunately, I had a replacement cable available so I could replace it; I had done this same fix several years ago and knew what was involved, but I had forgotten the fine points.

Replacing the cable is pretty quick and easy, less than five minutes -- but adjusting it so it shifts properly among the three front chainwheels takes time. In principle, it should be simple because there are only three (or sometimes four) parameters to adjust: the installed cable length; the high-shift limit screw; and the low-shift limit screw; many bicycles also have a "fine" adjustment for the cable length on the shift-lever itself.

There is a well defined procedure and sequence for making the adjustments to get the proper shifting performance, and there are many online videos describing it. Some are brief, around five minutes, but a few address the reality and run 30 minutes. The reason is that after the first pass through, you have to go back, tweak the screws, try again, tweak again, and so on: It's an iterative prices with interactive tweaks. Finding the precise combination of screw positions to get it all working right is time-consuming and aggravating, as improving one part of the derailleur's shifting usually degrades another part. (It's complicated by the fact that you want the shifter cage to overshoot slightly to ensure the chain snaps into position, and then settle back a little.

I don't fault the design of the derailleur -- it's a refined, elegant design that meets many conflicting objectives amazingly well. Certainly, it's not just mechanical designs that suffer from this malady. Many years ago, while involved in the design and implementation of an analog front end, we had a very senior designer on the team whose hallmark was elegance in circuitry. His designs almost always worked the first time (very impressive), and he also strived to minimize the number of parts on the BOM.

In this design, he got the number of mechanical trimpots down to about five, but in doing so, their adjustment was interactive (Figure 1) and recursive. While the whole BOM was shorter and less expensive, it was a time-consuming hassle to adjust the resultant circuit at the prototype stage to meet specs, and even more so for the production people doing the pilot run.

Pretty soon, the very legitimate complaints from the manufacturing folks were heard loud and clear. Management assigned an experienced and pragmatic engineer to redesign the front end. His design rearranged the circuit and needed two more trimpots, but none of their adjustments interacted (Figure 2). You could do it all in a couple of minutes, just by following a specific linear, non-recursive sequence. Problem solved, at modest cost in components.

Figure 2: When adjustments don't interact, calibration requires only one pass through the steps.

 

Now, of course, the interactive calibration and trim situation would be an entirely different story entirely. If physical circuit trimming were needed, designers would likely use digital pots (digitpots) that could be calibrated by an algorithm in a test fixture. Even if there were interrelated trims, the software could sweep through all the combinations to find the one that worked. Even more likely, the entire trim and calibration process would be done using software that would measure and record desired versus actual results at various points of the circuit's range, then put calibration factors or equations into memory to be used by the operational software as needed.

Have you ever been involved with the designing a circuit or function with interactive adjustments requiring integrative calibration steps? Was this an unavoidable situation, or was it due to lack of insight or concern? Or have you ever had to deal with a design that had this problem?